E&S Journal logo

Article

September 15, 2026

Credit Card Fraud Detection Using Neural Networks in Imbalanced Data Environments

Jefferson de Oliveira Ferreira; Lauro Marques Vicari

DOI: 10.22167/2675-6528-202602232

Article derived from a Final Course Work (TCC), with content based on the student’s original work and adapted to the editorial format of the Revista E&S with the support of the ResumeAI tool, an artificial intelligence solution developed by the Pecege Institute for textual synthesis and organization.

Summary

The growing digitalization of the financial system has made credit card transaction fraud a critical obstacle to institutional stability and customer trust. This study compared a Multilayer Perceptron (MLP) Neural Network architecture with a Logistic Regression model, used as a baseline, for fraud detection in a scenario of extreme data imbalance. A dataset of real credit card transactions was used, applying synthetic oversampling techniques (SMOTE) and the Focal Loss function to mitigate the bias of the majority class. The MLP was implemented with four dense layers, and Logistic Regression served as a reference. The experiments revealed that the Neural Network outperformed the logistic model in business metrics, achieving a Recall of 0.8553 and an F1-Score of 0.8483. Notably, the MLP reduced undue blocks (False Positives) by 23.08% compared to Logistic Regression, presenting itself as a more balanced solution between operational efficiency and user experience. However, both models faced difficulties in identifying high-value frauds due to the behavioral mimicry of the offenders. It was concluded that Neural Networks constitute a superior strategic choice for institutions seeking scalability and lower operational friction, although they require complementary business rules for high-value frauds.

Keywords: Credit Card; data science; Fraud Detection; Neural Networks; Logistic Regression.

1. Introduction

The payment ecosystem in Brazil has undergone a significant structural transformation in the last decade, driven by the widespread digitalization of financial services and the consolidation of e-commerce. In this scenario, credit cards have consolidated as the main transaction vector, registering unprecedented financial volumes, as pointed out by ABECS (2024). The migration to “Card Not Present” (CNP) modalities has brought a considerable challenge for operators, who face the refinement of digital fraud, operating in regimes of scale and speed that surpass conventional defense mechanisms.

In the national context, digital fraud has evolved into the “Fraud-as-a-Service” paradigm. Data from the Serasa Experian Indicator (2025) reveal that approximately 25% of fraud attempts detected in Brazil are related to banks and issuers, with an occurrence every 4.2 seconds. The ClearSale Fraud Map (2025) indicates that the average “ticket” for fraudulent transactions (R$ 1,083.00) is significantly higher than for legitimate transactions (R$ 549.00). This scenario compromises the viability of retail due to “chargeback”, a process in which the retailer absorbs the full loss, in addition to generating sanctions from acquirers.

Historically, fraud detection relied on traditional barriers, such as static rules and extensive manual reviews conducted by human analysts. However, these approaches lost effectiveness as the volume of transactional data grew exponentially and the sophistication of criminals increased. Fraudsters began to mimic the consumption behavior of legitimate customers to bypass previously fixed parameters, making the operational cost of maintaining large manual review teams unsustainable.

To overcome this inflexible logic, organizations have begun to implement artificial intelligence architectures capable of processing multiple variables dynamically. However, the transition to predictive models encounters challenges inherent in the nature of financial data, such as concept drift, a statistical phenomenon in which attack properties and patterns change rapidly over time (Dal Pozzolo, 2015). The dynamic nature of transactions requires algorithms that learn continuously and robust techniques to deal with extreme historical imbalance, where fraud is always a rare event in a vast ocean of legitimate data.

Related research demonstrates the analytical superiority of nonlinear models in overcoming the challenges inherent in financial data. Gouvêa et al. (2013) showed that neural architectures have superior aptitude for deciphering complex correlations and optimizing loss mitigation. Jordan and Mitchell (2015) highlight machine learning as a fundamental advancement for cybersecurity, enabling proactive threat prediction. Aggarwal (2018) and Schmidhuber (2015) explain that “Deep Learning” excels at hierarchically extracting latent, nonlinear patterns using multiple layers of abstraction. This flexibility is crucial for the precise calibration of loss functions and decision thresholds, allowing the institution to prioritize strategic security metrics, such as sensitivity (“Recall”), mitigating the drop in precision that penalizes traditional models. A misclassification resulting in a false positive, such as the undue blocking of a card, causes immediate operational friction, direct revenue loss, increased customer churn rate, and damage to the institution’s reputation, exceeding the nominal value of the chargeback the algorithm attempted to avoid.

Thus, the general objective of the present article is to develop and implement a predictive fraud detection model anchored in an “Artificial Neural Networks” architecture of the “Multilayer Perceptron” (MLP) type, and to compare it directly with a traditional Logistic Regression model (used as a “baseline”). The research seeks to validate the technical and operational effectiveness of both approaches on a representative experimental “dataset”, evaluating which algorithm delivers the best solution for addressing imbalanced data and for the strategic balance between security and consumer experience.

2. Material and Methods

For this project, a practical and data-driven (quantitative) approach was chosen. The objective was to structure a fraud detection process using deep neural networks and specific techniques to handle the scarcity of fraud observations. The code was developed in Python, employing Pandas for data manipulation and TensorFlow and Keras for modeling.

Data Source and Development Environment

The study used a “dataset” of real credit card financial transactions, obtained from the “Kaggle” repository, made available by the Machine Learning Group of Université Libre de Bruxelles and based on the studies by Dal Pozzolo (2015). The database comprises 284,807 transactions from European customers over two days. The dataset presented an extreme imbalance, with only 492 transactions classified as fraud, representing 0.172% of the total volume.

The original variables were not disclosed due to banking confidentiality regulations. In their place, 28 principal components (“features” V1 to V28) generated by a PCA (“Principal Component Analysis”) transformation were provided, with only the numerical variables Time (“Time”) and Amount (“Amount”) remaining in raw format. The presence of 28 components to represent the original variables indicates that the PCA technique was primarily employed as a cryptographic mechanism, which imposes a methodological fragility due to the absence of metadata about the original predictors, limiting the model’s explainability.

Exploratory Data Analysis and Pre-processing

An Exploratory Data Analysis (EDA) was performed using the Pandas library, confirming the integrity of the dataset and the absence of missing values. The severe data imbalance (0.172% of fraud) required a modeling strategy that went beyond the conventional, as biased algorithms would tend to ignore the minority class. For a robust analysis in extreme imbalance scenarios, metrics such as precision and recall were considered essential.

To mitigate data bias, the SMOTE (“Synthetic Minority Over-sampling Technique”) “oversampling” resampling technique was employed. SMOTE acts synthetically, identifying close neighbors of the minority class and creating new examples to balance class distribution. This technique was applied exclusively to the training set, after the separation of test data, to avoid “data leakage”. Before SMOTE, the training set contained 199,020 legitimate samples and 344 fraudulent ones. After rebalancing, the training set was equalized to 50% of each class, totaling 398,040 observations.

Table 1. Distribution of classes in the pre and post-SMOTE training set.

Processing stepLegitimate transactions (class 0)Fraudulent transactions (class 1)Total of samples
Original (pre-SMOTE)199.020344199.364
Balanced (post_SMOTE)199.020199.020398.040

Source: Original survey data

The disparity between the variables “Time” and “Amount” and the presence of “outliers” were treated with “StandardScaler”, which resizes the data to a common scale, centering the mean at zero and adjusting the standard deviation to the unit, according to Equation (1):

Z = (χ – μ) / σ (1)

where x represents the original value, μ the sample mean, and σ the standard deviation. This standardization ensures that the algorithm processes all “features” with the same hierarchical importance.

The database was subdivided into three sets: Training (70%), Validation (15%), and Test (15%). The training set was used for the convergence of synaptic weights, the validation set for monitoring performance and adjusting hyperparameters, and the test set, kept isolated, for the definitive evaluation of the model’s performance in a real-world scenario. A stratified split was used to preserve the original proportion of 0.172% of fraud in all subsamples (Training, Validation, and Test), ensuring the model’s reliability under real conditions.

Architecture and Functioning of the Neural Model

A “Multilayer Perceptron” (MLP) type Neural Network was implemented with four dense layers: an input layer with 30 neurons, three hidden layers (with 128, 64, and 32 neurons, respectively), and an output layer, according to figure 1.

Figure 1. Schematic representation of the proposed MLP Neural Network architecture

Source: Original survey data

Each neuron functions as a node that receives input signals, applies weights, and generates an activation. Dense layers are fully connected, allowing the model to learn global combinations and complex patterns. The internal functioning of each neuron is based on a linear transformation followed by a non-linear activation. The weighted sum of inputs (x) and weights (w), plus the bias (b), is calculated by:

z = (x .w) + b (2)

In the hidden layers, the ReLU (“Rectified Linear Unit”) function was applied to learn complex and non-linear patterns, accelerating the model’s convergence:

f(z) = max(0, z) (3)

To prevent “overfitting”, the hidden layers used L2 Regularization (λ = 0.0001), which adds a penalty to the loss function proportional to the square of the weights. In the output layer, the Sigmoid function was used, which compresses the result into a probability between 0 and 1:

σ(z) = 1 / (1 + e^(-z)) (4)

The learning process is based on the use of the error generated at the output as a guide for recalibrating the internal weights, through the “backpropagation” algorithm. This iterative cycle, known as the “training loop”, seeks to progressively minimize the cost function, confronting the model’s predictions with the actual targets (“labels”) to adjust the network based on gradient descent.

Figure 2. Neural network learning process

Source: Chollet (2011)

Hyperparameter tuning and experimentation

The final configuration of the neural network (128, 64, and 32 neurons) resulted from an iterative experimentation process. Table 2 presents the sensitivity metrics (“Recall”) of different architectures under the same training conditions and a “Business Approach” threshold (maximum of 50 False Positives).

Table 2. Sensitivity metrics (“Recall”) for different architectures.

Initial architectureStatistical RecallRecall BusinessDiagnosis
256 neurons86,49%81,08%Overfitting / Noise Memorization
16 neurons86,49%81,08%Underfitting / Low separation capacity
128 neurons86,49%85,14%Optimal point of generalization

Source: Original survey data

The selected intermediate architecture (128, 64, 32 neurons) proved to be the most robust, raising business Recall to 85.14% and maximizing the identification of fraudulent transactions without compromising learning stability. The Adam (“Adaptive Moment Estimation”) algorithm was chosen as the optimizer due to its high efficiency in scenarios with sparse data and complex loss spaces, calculating adaptive learning rates for each weight individually and ensuring faster and more stable convergence. The “Early Stopping” technique was incorporated to monitor the loss on the validation set (val_loss) and interrupt training when performance gain stagnated, combating “overfitting”.

Optimization

The “Focal Loss” technique was implemented as the cost function, replacing the standard binary cross-entropy. This metric is designed to handle extreme class imbalance by focusing learning on “hard” examples (rare frauds) and penalizing the influence of legitimate transactions. The function is defined by:

FL(pt) = −αt(1 − pt)γlog(pt) (5)

where (pt) is the estimated probability for the true class, αt is the balancing factor and γ (“focusing parameter”) modulates the loss of already well-classified examples. In this study, γ = 2 and α = 0.25 were adopted, defined statically based on the literature for unbalanced data. The error was minimized through backpropagation, guided by the Adam optimizer.

A methodological differentiator was the flexibility of the classification threshold. A systematic search routine (“Threshold Moving”) was applied to identify the optimal operating point of the model, aligned with business criteria and risk appetite, and not just statistical maximization. The threshold was adjusted to keep the volume of undetected fraud (False Negatives) below 14 cases in the test set, prioritizing the containment of direct loss and minimizing False Positives.

Reference models and comparative foundation

For validation, the MLP architecture was compared with Logistic Regression, adopted as a “baseline” due to its history in “credit scoring” and fraud prevention systems. Logistic Regression attempts to separate the classes through a linear combination of the variables, mapped to a probabilistic interval by the sigmoid function:

P(y = 1|x) = 1 / (1 + e^(-(β0+β1×1+…+βnxn))) (6)

where P(y = 1|x) is the probability of a fraudulent transaction, β0 is the intercept and βi are the coefficients adjusted for each variable xi. The comparison aimed to verify if the Neural Network’s complexity would result in a real gain in “Recall” and a reduction in False Positives, and if the advanced model would extract latent information ignored by linear models. For fairness, Logistic Regression was also subjected to the threshold moving routine to achieve the same security rigor stipulated for the main model, allowing for a paired comparison of operational cost.

Use of AI in work development

During the preparation of the work, Gemini (Google) and ChatGPT (OpenAI) were used to assist in the literature search, methodology alignment, review, formatting, and translation of the abstract. The content was reviewed and edited by the author, who assumes full responsibility.

3. Results and Discussion

This section presents the results obtained by the Neural Network architecture compared to the baseline model (Logistic Regression). The discussion is not limited to statistical rigor, but seeks to translate performance metrics into business impact, evaluating the balance between security and operability.

Descriptive data analysis

Initially, the distribution of transaction values was analyzed (Figure 3). To enable density visualization and mitigate the effect of extreme outliers, transactions up to $2,000 were selected, a range that comprises approximately 99.76% of the total database. It was observed that both legitimate and fraudulent transactions are concentrated in low values, however, fraud shows a slightly higher density in microtransactions, possibly indicating card testing.

Figure 3. Probabilistic Density of Values

Source: Original research results

When converting the original time to a 24-hour cycle (Figure 4), it was observed that legitimate transactions follow the standard commercial rhythm. In contrast, fraud maintains a relevant density during the early morning hours (between 2 AM and 6 AM), a period of lower active user surveillance, which characterizes the temporal factor as a strong indicator of anomaly. The proportional increase in attempts during lower transactional volume hours suggests an attempt to exploit windows with lower monitoring sensitivity. Therefore, the inclusion of this feature offers the neural network a valuable contextual dimension to differentiate typical from atypical behavior.

Figure 4. Transaction density per hour

Source: Original research results

Continuing the relevance analysis, Figure 5 presents the association between latent variables (V1 to V28) and the occurrence of fraud, using the point-biserial correlation coefficient. The mapping revealed that variables such as V17 (negative) and V11 (positive) are the main indicators of fraud. Variables like V17, V14, and V12 exhibit more pronounced negative associations, indicating that lower values in these dimensions increase the probability of fraudulent transactions. On the other hand, V11 and V4 show the largest positive associations, acting as risk indicators as their values increase. The density distributions of the variables with the greatest impact were examined individually, visually confirming the separation between classes.

Figure 5. Correlation between the variables

Source: Original research results

In Figure 6, the behavior of the variable V11 (positive correlation) is observed, where the displacement of the fraud data mass (in red) to the right of the normal transaction curve is clear, ratifying V11 as a robust predictor. Similarly, Figure 7 details the variable V17 (negative correlation), where the density of frauds is severely concentrated in the left tail (negative values), as predicted by the point-biserial coefficient. The differential of the MLP Neural Network lies in its ability to process these polarities in a multivariate manner, capturing complex fraud patterns that escape simple linear association metrics.

Figure 6. V11 density distribution

Source: Original research results

Figure 7. V17 density distribution

Source: Original research results

Monitoring of Metrics and Identification of “Overfitting”

The neural network performance evaluation began with monitoring the learning curves over the training epochs, as detailed in Figures 8 and 9. The results revealed a classic scenario of early overfitting. In the Loss graph, it was observed that while the training error decreased steadily, the validation curve showed an upward and unstable inflection after the first epoch, indicating that the model began to adjust excessively to the noise in the training set. This behavior was corroborated by the evolution of the AUC-ROC metric, where the model quickly reached the maximum score (AUC-ROC = 1.00) on the training set but failed to maintain stability on the validation data, oscillating between 0.91 and 0.94. This disparity between the curves confirmed that, despite the L2 regularization techniques, the network tended to memorize historical patterns at the expense of generalization. Therefore, interrupting the training via Early Stopping proved fundamental to preserving the artifact’s practical utility, triggering the stop at the 6th epoch to ensure the model ceased learning at its point of greatest balance.

Figure 8. Evolution of loss

Source: Original research results

Figure 9. Evolution of AUC (area under curve) during training

Source: Original research results

Comparative analysis of performance metrics

To validate the effectiveness of the proposed architecture, a direct comparison was made between the MLP Neural Network and the Logistic Regression baseline. Although the logistic model presented solid performance for evident transactional patterns, the Neural Network demonstrated a refined superiority by capturing non-linear nuances that escape traditional regression. Table 3 consolidates this technical disparity.

Table 3. Comparison of metrics

MetricLogistic RegressionNeural Network (MLP)Variation (%)Variation (p.p)
“Recall” (sensitivity)0,83780,85141,60+ 1,36
Precision0,65960,70796,82+ 4,83
F1-Score (Balance)0,73810,77304,51+ 3,49
AUC-ROC (Global)0,96580,9584-0,77– 0,74
False Negatives (FN)1211-9,09– 1
Improper Lockouts (FP)3226-23,08– 6

Source: Original research results

The most relevant data for the financial operation is the Recall of 0.8514 achieved by the Neural Network. In the context of payment methods, Recall is the sovereign metric, as it reflects the system’s ability to minimize False Negatives. The 9.09% reduction in missed fraud (from 12 to 11 cases in the test) represents direct savings in chargeback costs. In parallel, the significant drop of 23.08% in undue blocks (False Positives) is a critical gain in user experience, avoiding unnecessary friction with legitimate customers.

The class separation capability is visualized in the comparison of the ROC Curves (Figure 10). Both trajectories exhibit an ideal proximity to the upper-left corner of the graph, attesting to the high quality of the developed models. Although Logistic Regression boasts a slightly superior global AUC-ROC metric (0.9658), this value reflects a theoretical average performance encompassing all possible decision thresholds, including extreme cutoff points that would never be adopted in a real-world scenario. In the region of practical business interest, where the goal is to maximize fraud detection with the least possible impact on undue blocks, the Neural Network trajectory shows itself to be equivalent and, in certain intervals, more stable. This stability means that the model can sustain a high hit rate without suffering a disproportionate and abrupt increase in false alarms, confirming that the Deep Learning-based architecture is highly competitive and robust for decoding the complexity of latent variables in the credit card ecosystem.

Figure 10. ROC Curve Comparison: Logistic Regression vs. MLP Neural Network

Source: Original research results

Operational performance and threshold optimization

To evaluate the model by simulating the production environment, the confusion matrix was applied to the test base, ensuring the impartiality of the predictions. In scenarios of extreme imbalance, the analysis of this matrix is indispensable for quantifying the exact volume of correct predictions (True Positives) against the critical financial risk of undetected fraud (False Negatives). The high class separation capacity demonstrates the success of the adopted hybrid approach. The SMOTE algorithm intervened in the preprocessing, mitigating the deficit of the base by synthesizing new samples for the minority class. In contrast, Focal Loss acted directly on the neural network training, calibrating the cost function to prioritize borderline and difficult-to-classify fraud. This dual strategy enabled the construction of a robust decision boundary. The visual result of this strategy can be observed in Figure 11, which presents the comparison of the confusion matrices. It was noted that the MLP achieved a significant reduction in False Positives compared to the logistic baseline, maintaining an equivalent level of security in detecting real fraud.

Figure 11. Confusion matrices – Test set

Source: Original research results

A central point of this stage was the rejection of the standard statistical threshold of 0.5. As illustrated in Figure 12, the strategic calibration of the trade-off between Precision and Sensitivity (Recall) allowed for the identification of the optimal operating point for the real-world scenario. This decision prioritized operational sustainability, establishing a strict ceiling for the volume of False Positives to avoid undue customer blocking, in exchange for rigorous maintenance of high-impact fraud capture. Figure 12 illustrates the fundamental dilemma between Precision (Blocking Quality) and Recall (Capture Power) in statistical and business approaches. In the purely statistical approach, both models aim for maximum capture (Recall), resulting in high fraud detection (LR = 0.91; MLP = 0.86), but with unacceptably low precision (LR = 0.02; MLP = 0.07), generating an unsustainable volume of false positives. By adopting a business perspective, decision thresholds are adjusted to balance fraud capture with operational cost and customer experience. The analysis of the graph reveals the operational superiority of the Neural Network model [MLP]. While the Logistic Regression model [LR] undergoes a severe trade-off to achieve an operational Precision of 0.66, significantly sacrificing Recall (dropping to 0.84), the MLP model demonstrates remarkable efficiency. The shift of the MLP to the business strategy achieves superior Precision (0.71) with almost no loss of Recall (dropping only from 0.86 to 0.85). This characteristic proves that the MLP model is more robust and capable of finding optimized operating points, maximizing fraud capture with minimal operational friction. This double gain validates the hypothesis that the Neural Network’s ability to process non-linear patterns through its hidden layers allows for a more refined separation in the decision space, overcoming the intrinsic limitations of linear models in complex banking data.

Figure 12. The Decision Displacement: Statistics x Business

Source: Original research results

Statistical analysis of errors and value disparity

To deepen the discussion on the model’s effectiveness, a comparative analysis was conducted between detected frauds and those the model failed to identify. Reversing the mathematical standardization allowed observation of the real transaction values in their original magnitudes. The analysis revealed a critical behavioral pattern, as shown in Figures 13 and 14. In both models, lost frauds (False Negatives) presented a significantly higher median value than those successfully detected. In Logistic Regression, detections primarily occurred in low-value transactions (median of $2.00), while failures were concentrated in operations with a median of $231.64. In the Neural Network, the phenomenon was analogous: the model was effective in capturing small-value frauds but encountered greater difficulty in transactions with a median of $276.17. This scenario raises a warning about the model’s applicability: frauds that generate the largest financial losses are managing to bypass the detection system due to the behavioral mimicry of criminals who imitate the consumption of high-ticket legitimate customers. There is no significant financial difference between the architectures regarding the capture of major fraudsters, as both algorithms allow these transactions to pass. The true trade-off between the methodologies lies in operational efficiency with the customer base, where the Neural Network (MLP) acts more cleanly, generating 18.7% fewer undue blocks compared to the baseline.

Figure 13. Baseline Financial Impact: MLP

Source: Original research results

Figure 13. Baseline Financial Impact: Logistic Regression

Source: Original research results

The present investigation demonstrated that the Multilayer Perceptron (MLP) Neural Network architecture offers real gains over traditional Logistic Regression in credit card fraud detection, especially in imbalanced data environments. The strategic implementation of synthetic oversampling (SMOTE) in conjunction with the Focal Loss loss function optimized the model’s resistance to the majority class bias, resulting in high levels of F1-Score and Precision, as well as training stability that minimizes oscillations in generalization capacity. The MLP showed an 18.7% reduction in False Positives compared to Logistic Regression, translating into greater operational efficiency and a better user experience. However, error analysis revealed that the highest nominal value frauds still manage to bypass the neural network barrier due to the behavioral mimicry of criminals, indicating that the ability to predict the mathematical probability of an event must be complemented by mechanisms that weigh the absolute financial risk involved.

4. Conclusion

This study aimed to develop and compare a predictive model for credit card transaction fraud detection, using a Multilayer Perceptron (MLP) Neural Network architecture against a Logistic Regression baseline, focusing on handling imbalanced data and the strategic balance between security and user experience. It was found that the MLP demonstrated superiority in business metrics, achieving a Recall of 0.8514 and a 23.08% reduction in undue blocks (False Positives) compared to Logistic Regression. The strategic application of synthetic oversampling techniques (SMOTE) and the Focal Loss loss function was fundamental to mitigating the bias of the majority class and optimizing model robustness. This approach resulted in high levels of F1-Score and Precision, as well as training stability that minimized oscillations in generalization capability. The Neural Network, therefore, constitutes a superior strategic choice for financial institutions aiming for scalability and reduced operational friction, without compromising security.

However, the error analysis revealed that both models faced difficulties in identifying high nominal value frauds, attributed to the behavioral mimicry of criminals, who manage to bypass the detection network. This limitation indicates the need to complement mathematical predictive capacity with mechanisms that weigh the absolute financial risk involved. Additionally, the use of a static dataset restricted the view on the dynamic and temporal nature of frauds, and the PCA transformation imposed a fragility in the causal explainability of the original variables. As future developments, it is suggested to implement secondary business rules focused on containing high-value anomalies and transitioning to recurrent architectures, such as LSTMs or RNNs, capable of interpreting the cardholder’s history as a logical and chronological sequence, improving detection in “Concept Drift” scenarios.

Bibliographic References

ABECS. 2025. Balanço do setor de meios eletrônicos de pagamento: resultados 2024. Associação Brasileira das Empresas de Cartões de Crédito e Serviços, São Paulo, SP, Brasil. Disponível em: Acesso em: 05 fev. 2026.

Aggarwal, C.C. 2018. Neural Networks and Deep Learning. Springer, Cham, Suíça. Disponível em: . Acesso em: 12 out. 2025.

Cholet, F. 2021. Deep learning with python. 2ed. Manning, New York, EUA.

ClearSale. 2025. Mapa da Fraude. Disponível em: Acesso em: 10 jan. 2026.

Dal Pozzolo, A. 2015. Adaptive Machine learning for credit card fraud detection. Tese, Ph.D. em Ciência da Computação, Université Libre de Bruxelles. Disponível em < http://di.ulb.ac.be/map/adalpozz/pdf/Dalpozzolo2015PhD.pdf>

Gouvêa, M.A.; Gonçalves, E.B.; Mantovani, D.M.N. 2013. Análise de risco de crédito com aplicação de regressão logística e redes neurais. Contabilidade Vista & Revista 24(4): 96-123. Disponível em: < https://www.redalyc.org/pdf/1970/197033497006.pdf > Acesso em: 12 out. 2025.

Jordan, M.I.; Mitchell, T.M. 2015. Machine learning: Trends, perspectives, and prospects. Science 349(6245): 255-260. Disponível em: Acesso em: 12 out. 2025.

Schmidhuber, J. 2015. Deep learning in neural networks: An overview. Neural Networks 61(1): 85-117. Disponível em: . Acesso em: 15 out. 2025.

Serasa Experian. 2025. Record: Almost 7 million fraud attempts were registered in the 1st quarter of 2025; the banking sector is the main target. Available at: Accessed on Jan. 10, 2026.

Article originating from the Course Conclusion Work of the Specialization in Data Science and Analytics of the MBA USP/Esalq

To learn more about the course, click here and access the MBX Academy platform

Most recent

October 22, 2026

Challenges and Possibilities of the First School Year of Children with ASD for Teaching Work

October 22, 2026

Consumer Journey in the Decision to Purchase Infoproducts in the Digital Environment

October 22, 2026

Psychosocial Risks and Mental Health at Work in a Banking Institution

October 22, 2026

Digital Observatory of the Amazon: Data Science Applied to the Analysis of Social, Economic, Environmental, and Health Indicators

October 22, 2026

Continuous Learning as a Factor of Organizational Growth in the Perception of Collaborators

October 22, 2026

Statistical Regression in Microgrids for Operational Performance Estimation: Campusgrid Case

October 22, 2026

Forecast of Aggregate Default in the Brazilian Financial System by Sarimax Model with Macroeconomic Variables

October 22, 2026

Stakeholder Engagement Factors in International Technology Transfer to a Public Hospital

October 22, 2026

Combating Commercial Electric Energy Losses Using a Time Series Model

You may also like

October 22, 2026

Challenges and Possibilities of the First School Year of Children with ASD for Teaching Work

The inclusion of children with Autism Spectrum Disorder (ASD) in Early Childhood Education presents significant challenges, given the difficulty of adapting to new environments and the intensity of stimuli. The objective was to analyze the teaching performance in Early Childhood Education in the initial schooling process of children with ASD. A case study was conducted, using narrative interview as a data production technique. The results revealed that the teaching performance was characterized by mediation and affection practices, with the flexibility of routine and the use of playful resources as essential strategies for the child’s development and sensory organization. The importance of the school support professional for the success of the pedagogical work was also highlighted. The child’s integral development was promoted through intentional teaching actions that prioritized environmental adaptation, strengthening of bonds, and attention to the specific needs of ASD. As a contribution, a guide of pedagogical strategies aimed at welcoming children with ASD in public daycare centers was developed.

Keywords: Early Childhood Education; School Inclusion; Teacher Narratives; Inclusive Pedagogical Practices; Autism Spectrum Disorder.

October 22, 2026

Consumer Journey in the Decision to Purchase Infoproducts in the Digital Environment

The expansion of e-commerce and technological transformation have driven the need to understand consumer behavior regarding infoproducts, whose perceived experience has become crucial for professional development. The study aimed to analyze the factors that influenced the purchase decision of infoproducts in the digital environment, considering the stages of the consumer journey. An applied, quantitative, and descriptive research was conducted through a field survey with 100 consumers from Serra Gaúcha. Data collection occurred via a structured questionnaire, using non-probabilistic sampling. The results revealed a qualified audience with strong “mobile-first” behavior. It was found that the availability of detailed information and social approval were the main attraction triggers in the pre-purchase phase. In the transactional and subsequent stages, the minimization of operational friction, the agility of immediate access, and system security proved crucial for conversion and satisfaction, although gaps in human support efficiency were identified. Retention was based on the trust generated by positive past experiences, overcoming the price factor. It was concluded that the infoproduct consumer journey operates as an integrated system, requiring alignment between technological infrastructure, content curation, and support humanization to ensure engagement.

Keywords: E-commerce; Consumer behavior; Consumer experience; Loyalty; Virtual business.

October 22, 2026

Psychosocial Risks and Mental Health at Work in a Banking Institution

Mental health in the workplace is a topic of growing relevance, impacting professional development and employee well-being. A study was conducted to identify aspects and situations that can influence absenteeism levels in a banking institution. The research involved 108 participants and used a structured questionnaire, based on the short version of COPSOQ II by Gonçalves (2021), which contained 45 closed Likert scale questions and two open-ended questions. Data were collected via an online platform and analyzed with a descriptive and quantitative approach. The results revealed a scenario that indicated the need for adjustments in the distribution of activities and in managers’ perception of behaviors and signs of disorders. Important evidence was identified in the participants’ open-ended responses. It was concluded that the study contributed to promoting the perception of a safe environment for mental health in the organization.

Keywords: Absences due to disorders; Professional development; Operational leadership; Medical leave; Mental health.

October 22, 2026

Digital Observatory of the Amazon: Data Science Applied to the Analysis of Social, Economic, Environmental, and Health Indicators

The state of Amazonas faced historical challenges in consolidating the Unified Health System (SUS), stemming from geographical barriers, socioeconomic inequalities, and structural limitations. This study described the development and implementation of a health information geoportal, designed to support monitoring, evaluation, and planning in the Amazonian context. A descriptive case study approach was adopted, based on documentary analysis and secondary data, covering the mapping of functionalities, technical infrastructure, and validation steps of the tool. The Portal was developed with open technologies and integrated multiple national databases, offering interactive dashboards, thematic maps, and a digital document repository. Among the results, highlighted were the georeferenced visualization of key indicators, the possibility of temporal comparative analyses, and the use of the platform in reviewing health plans and redirecting strategic actions. The portal can represent an advance in digital governance and evidence-based management, with potential for replication in other highly complex territories, provided it is accompanied by investments in technological infrastructure, professional training, and inter-institutional coordination.

Keywords: Amazon; Health Evaluation; Digital Health; Health Information Systems.

October 22, 2026

Continuous Learning as a Factor of Organizational Growth in the Perception of Collaborators

The impact of continuous learning and knowledge management on organizational growth was studied in a context of globalization and constant changes. The study’s objective was to analyze employees’ perception of learning culture and its importance for personal and professional development. Data were obtained through an anonymous survey, with a quantitative approach, using a form with “survey” type questions applied to 87 professionals from the technology, health, education, commerce, and services sectors. The results indicated that employees had a positive perception of the contribution of continuous learning to personal and organizational growth, recognizing its benefits for the development of new skills, innovation, and competitiveness. However, it was observed that the incentive for internal knowledge exchange and the offering of practical training still presented moderate levels of agreement. It was concluded that continuous learning significantly contributes to individual and organizational development, suggesting that companies should develop strategic policies for learning focused on new skills, aiming for greater competitive advantage and innovation.

Keywords: Continuous learning; Self-development; Organizational growth; Knowledge Management; Innovation.

October 22, 2026

Statistical Regression in Microgrids for Operational Performance Estimation: Campusgrid Case

A multiple statistical regression model was proposed to estimate the operational performance of UNICAMP’s CampusGrid microgrid, based on internal electrical quantities. The microgrid integrates photovoltaic generation, batteries, and a gas generator. Energy management system data were collected at five-minute intervals over one month. The Operational Reliability Index, defined as the ratio between supplied and demanded energy, was established as the dependent variable. The model was adjusted for the islanded operation mode, evaluating the impact of demand, state of charge, solar generation, and supplied energy. The results revealed high explanatory power, indicating that operational reliability depends on the interaction between supply and demand. It was observed that the nighttime period concentrated the highest rate of partial supply, attributed to the reduction in solar generation. It was concluded that multiple regression offers an interpretable and useful framework for microgrid monitoring and control strategy support.

Keywords: Battery storage; Photovoltaic generation; Reliability Index; Islanded operation; Energy Management System.

October 22, 2026

Forecast of Aggregate Default in the Brazilian Financial System by Sarimax Model with Macroeconomic Variables

Aggregate non-payment in the Brazilian financial system is a crucial indicator of credit risk, the prediction of which is regulated by CMN Resolutions n.° 4.557/2017 and 4.966/2021 for the prospective measurement of expected losses. Aggregate non-payment was modeled and predicted using the SARIMAX model, projecting its trajectory until December 2026 under optimistic, base, and pessimistic scenarios. 164 monthly observations (July/2012 to February/2026) and macroeconomic variables such as unemployment, Selic, IPCA, IBC-Br, and household debt were used. The specification with contemporaneous variables was selected based on information criteria and coefficient stability, and residual diagnostics confirmed statistical adequacy, with a MAPE of 3.12%. Historical non-payment ranged between 4.33% and 7.64%. The estimation identified unemployment (β = 0.2534; p < 0.001) and Selic (β = 0.2128; p < 0.001) as the main positive determinants, with strong temporal persistence (AR(1) = 0.8965). Selic showed increasing correlation with non-payment (r = 0.62 at lag 12), while unemployment demonstrated a contemporaneous effect (r = 0.33). Projections indicated non-payment between 6.30% and 7.98% in December 2026, with a range of 1.68 percentage points, highlighting direct implications for provisioning in financial institutions.

Keywords: Unemployment; Predictive modeling; Monetary policy; Loss provisioning; Time series.

October 22, 2026

Stakeholder Engagement Factors in International Technology Transfer to a Public Hospital

This study described the critical factors perceived for stakeholder engagement in international technology transfer in a public hospital, with a focus on planning. The objective was to describe these conditioning factors, according to the perception of professionals. A single case study was conducted, applying a questionnaire to 25 professionals and using content analysis, complemented by descriptive tabulations. The main obstacles identified were resistance to change (72%), funding (64%), lack of communication (60%), and bureaucracy (40%). Only 28% of respondents recognized the existence of adequate tools for stakeholder management, while 48% denied it and 24% did not know. Exploratory cross-tabulations indicated variations in the assessment of influence according to time of practice, but did not establish causal relationships. Lack of engagement was associated with delays, low adherence, waste of resources, communication failures, compromised knowledge absorption, and risks to patient safety. The conclusions were limited to the perceptions of the researched group and supported preliminary recommendations for intersectoral communication and coordination.

Keywords: Case study; Project Management; Health innovation; Stakeholders; Public sector.

October 22, 2026

Combating Commercial Electric Energy Losses Using a Time Series Model

Commercial losses of electric energy represented one of the main challenges for distributors, directly impacting economic sustainability and supply quality. The efficiency of time series models, focusing on exponential smoothing techniques, was evaluated in forecasting billed consumption of consumer units, to support the identification of deviations associated with non-technical losses. Historical data of monthly consumption from an electric energy distributor in Brazil were used, segmented by behavior profiles (stationarity, trend, and seasonality). Simple exponential smoothing models, Holt, and Holt-Winters (additive and multiplicative) were applied, compared to a traditional moving average. The Mean Absolute Percentage Error (MAPE) was the evaluation criterion. Additionally, the ETS approach was employed for automatic model selection and residual diagnostic tests, such as Ljung-Box and autocorrelation function analysis. The results demonstrated that more robust models showed a greater capacity for adaptation to the series’ characteristics and lower error, with a highlight on superior performance in scenarios with trend. For seasonal series, the ETS model showed a greater capacity for adaptation to transitions. It was concluded that the use of more structured time series models contributed significantly to the improvement of forecast accuracy and offered relevant analytical support for the identification of commercial losses and the optimization of operational and commercial processes in the electric sector.

Keywords: Energy distribution; MAPE; Optimization; Consumption forecasting; Exponential smoothing.

Economy

October 19, 2026

End of the 6×1 scale: what do the numbers allow us to conclude?

Following the promulgation of the end of the 6×1 shift system, the debate on the shorter workday moves to implementation and requires distinguishing social benefits from economic costs. The end of the 6×1 shift system transitioned from legislative discussion to a constitutional change in Brazil. On October 16, 2026, the National Congress promulgated Constitutional Amendment No. 141/2026, originating from Proposal […]

Artificial Intelligence

October 16, 2026

Why study if AI exists?

When answers become easy, learning comes to mean more than just arriving at them. Why do I need to learn this if I can ask an artificial intelligence? This type of question has accompanied education for decades. Why learn an operation if a calculator exists? Why memorize information if it can be found […]

Tax Management

October 15, 2026

Pink tax and the gender of price

How different marketing treatments for men and women are incorporated by consumption taxation Imagine a shelf on which two razors of the same brand and intended for the same purpose are displayed side by side: one has dark packaging and promises strength, precision, and efficiency; the other comes in shades of pink or […]

Health

October 13, 2026

The missing link in health transformation

Why is innovation not enough to drive changes in organizations and health systems Digital transformation has become a strategic priority for health systems worldwide. Electronic health records, artificial intelligence, telemedicine, remote monitoring, and data analysis are now part of the strategic agenda of governments and health organizations. Paradoxically, […]

Marketing

October 09, 2026

Bitcoin or pets: what Artificial Intelligence would not see

A real-life consumer intelligence case shows what gets left out of the brief A few years ago, an American insurance company hired a specialized consumer intelligence firm to answer three questions: The insurer was also considering launching paid insurance in bitcoin. At the time, the consultancy’s workflow was as follows: the analyst created a query […]

School Management

October 06, 2026

The school is not a casino: when trading cards become currency in the playground

Why, for certain students, does school manage to offer so few experiences of achievement? The World Cup is over. Brazil left the competition, the school holidays came, the final was played and the football calendar moved on. It would be reasonable to imagine that, with the end of the tournament and the return to classes, […] would also begin

Energy Transition

September 30, 2026

Renewable energies: the Brazilian advantage

Brazil needs to invest in the capacity to transform natural resources into innovation, productivity, and development. Talking about energy is talking about development. Behind industry, transportation, services, cities, and increasingly artificial intelligence, there is an energy infrastructure that sustains the economy. Understanding a country’s energy matrix is understanding […]

Data Analysis

September 30, 2026

When the data changes, has reality changed?

An analysis of almost 20 million teaching positions-year between 2020 and 2025 shows how changes in population composition and record production can alter the interpretation of historical series Between 2020 and 2025, the number of formal teaching employment positions identified in RAIS (Annual Relation of Social Information), from the Ministry of Labor, […]

Artificial Intelligence

September 28, 2026

Artificial intelligence and the trends

Why do so many people join in on the fun with ChatGPT? I was never the trend person. I still am not. Some pass through my timeline, I see them, I find them interesting but not to the point of trying them out. This was the case, inclusive, with some recent waves of images made by artificial intelligence, like the 80s trend, the one that […]

Tax Management

September 01, 2026

Tax reform and estate planning in the countryside

Efficiency in managing taxes facilitates governance and allows for the organization of family succession. Rural wealth organization has become a strategic measure in light of the tax reform in Brazil, which came into effect this year. The planning of activities and rural wealth management, through tools such as the transition from an Individual to […]

Renata de Gaspari Valdejão Almeida

Artificial Intelligence

August 05, 2026

Inside the company and off the radar: the risks of Shadow AI

According to Microsoft, 75% of professionals already use artificial intelligence at work without company approval. The use of generative artificial intelligence is already part of the corporate routine. Tools like ChatGPT, Copilot, and other platforms help professionals produce texts, analyze data, create presentations, and accelerate tasks in practically all sectors. But this advance […]

Renata de Gaspari Valdejão Almeida

Strategy

July 02, 2026

The future as a competence

Academic interest reinforces the projection of scenarios as an essential tool for decision-making. While some companies are still trying to get the next step right, others are already working with multiple possible futures — and preparing for all of them. But after all, what does it mean to “think about the future”? Is it a trend, a strategy, or a new competence that […]

Renata de Gaspari Valdejão Almeida

Innovation

Technology

May 21, 2026

Agro SP Corridor: innovation, institutional articulation and future

Pecege commands the executive operation of the 400 km ecosystem activities that became known as the Brazilian agro Silicon Valley In 2025, in addition to taking over the management of the Piracicaba Technology Park, the Pecege Institute was also designated to command the executive operation of the State Agribusiness Innovation Corridor activities, the so-called […]

Education

April 27, 2026

How to turn a TCC into a business

New management of the Piracicaba Technology Park offers consulting to MBA graduates The new project director of the Piracicaba Technology Park (PTP), Alexandre Barreto, often says that one of the greatest rewards he has ever had in life was seeing his MBA USP/Esalq students structure companies from their final projects […]

Technology

March 20, 2026

The human repertoire in the age of AI

Critical thinking and creativity are irreplaceable in the work of a marketing professional Artificial Intelligence (AI) is capable of incredible feats, but it does not yet possess the judgment, empathy, intuition, and strategic creativity necessary to do the job of a good marketing professional. In this area, it is important to know how to use the tool to automate […]

Renata de Gaspari Valdejão Almeida

Agribusiness

Innovation

February 23, 2026

ESG as a growth strategy in a cooperative in RS

Santa Clara balances social and environmental responsibility with innovation Founded in 1911 by Italian immigrants in Carlos Barbosa, in the Serra Gaúcha region, Cooperativa Santa Clara is an example that it is possible to balance economic growth with social and environmental responsibility. From a small cheese production company, the organization has evolved into a […]

Renata de Gaspari Valdejão Almeida

Agribusiness

Innovation

January 22, 2026

Brazil free of Foot-and-Mouth Disease: what changes for Brazilian beef?

With a herd of over 200 million head of cattle, the country is currently the leader in exports. Brazil recently received, for the first time, the certification of country free of foot -and-mouth disease without vaccination, conferred by the World Organisation for Animal Health (WOAH). The seal strengthens the country’s position as a reliable supplier of meat for the […]

Renata de Gaspari Valdejão Almeida

Innovation

December 09, 2025

Innovation: the new cycle of the Piracicaba Technology Park

Instituto Pecege assumes management, betting on the creation of new businesses Starting in 2026, Instituto Pecege will assume the management of the Piracicaba Technology Park (PTP). Among its responsibilities will be the administration, operation, maintenance, and expansion of the park, in addition to fostering activities in innovation, entrepreneurship, and science and technology. The choice was made […]

Renata de Gaspari Valdejão Almeida

July 06, 2026

Large-Scale Architecture: DSR-SCRUM Change Management

Mônica Alessandra Guerios; Maria do Carmo Assis Todorov Abstract generated by ResumeAI, an artificial intelligence solution developed by Instituto Pecege focused on synthesis and writing. Architecture and urbanism are configured as essential bases in an increasingly complex and challenging urban space, where the intrinsic connection between social, environmental, and technological issues […]

Executive Mba In Leadership And Management

June 29, 2026

Supplier Qualification: Standardization and Risk

Isis Penelope Sumarelli Albuquerque Huang; Vítor Melão Cassânego Abstract generated by the ResumeAI tool, an artificial intelligence solution developed by Instituto Pecege focused on synthesis and writing. The growing complexity of agri-food supply chains, driven by globalization, has generated an intensified demand for transparency, security, and traceability in food systems. The integrity and authenticity […]

Executive Mba In Leadership And Management

June 26, 2026

Legal: from cost center to strategic partner

Adriana Ehiar Ribeiro Petry; Ana Carla Fernandes Gasques Abstract prepared by ResumeAI, an artificial intelligence solution developed by Instituto Pecege focused on synthesis and writing. The expansion scenario in the Brazilian market, although promising for consumer goods companies, presents significant challenges that require a strategic and adaptable approach. With a population […]

Executive Mba In Leadership And Management

June 26, 2026

Culture: Professionalization and growth in family businesses

Yashmin Meirelles Oliveira Araujo; Gilmar Martinelli Junior Abstract prepared by the ResumeAI tool, an artificial intelligence solution developed by Instituto Pecege focused on synthesis and writing. The Brazilian economic scenario is markedly influenced by the expressive representativeness of family businesses, which constitute approximately 90% of the country’s enterprises, according to PwC (2023) data. Within this vast universe, […]

Executive Mba In Leadership And Management

June 26, 2026

Semi-Fixed Teams: Engagement and Cohesion in the Field

Wesley Araújo Martins; Odair Silva Soares Abstract prepared by the ResumeAI tool, an artificial intelligence solution developed by Instituto Pecege focused on synthesis and writing. The operational sector of a company specialized in technical field services, operating in the maintenance of power transformers in industrial environments and energy utilities, faces significant challenges related to […]

Executive Mba In Leadership And Management

June 26, 2026

AI in pre-sales: Automated proposal generation

Vitor Luis Genaro; Felipe Carvalhal Abstract generated by the ResumeAI tool, an artificial intelligence solution developed by Instituto Pecege focused on synthesis and writing. The technology consulting market demands increasing operational efficiency from organizations, combined with analytical capacity and speed of customer response. In this dynamic scenario, the pre-sales area assumes […]

Executive Mba In Leadership And Management

June 26, 2026

PMaaS: Leveraging projects in SMEs without a PMO

Vitor Conceição Faria; Ewerton Mauro Visotto Faria Abstract elaborated by the ResumeAI tool, an artificial intelligence solution developed by Instituto Pecege focused on synthesis and writing. The contemporary business scenario, marked by high competitiveness and accelerated digital transformation, imposes complex challenges on companies, especially small and medium-sized ones. Many of these organizations, although engaged […]

Executive Mba In Leadership And Management

June 26, 2026

Agile Lifecycle Management in Public Funding

Vinícius Nascimento Garcia; Ana Carla Gasques Abstract generated by the ResumeAI tool, an artificial intelligence solution developed by Instituto Pecege focused on synthesis and writing. The state public development agency faced a scenario of reactive and fragmented development of its credit lines, characterized by low data usage and the absence of governance […]

Executive Mba In Leadership And Management

June 26, 2026

Mentoring and Onboarding: Developing Talent in GC

Vinícius Andrade de Carvalho; Ewerton Mauro Visotto Faria Abstract generated by ResumeAI, an artificial intelligence solution developed by Instituto Pecege focused on synthesis and writing. The demand for outsourcing services has experienced remarkable growth both in Brazil and globally, driving companies to expand their outsourced operations. Reports such as the “Global Outsourcing Survey […]