This study analyzed the relationship between brand collaborations, known as “collabs”, in the cosmetics sector and the production of user-generated content (UGC) on digital platforms. The research addressed the relevance of the topic given the growth of the beauty market and the intensification of “co-branding” strategies as a tool for differentiation and innovation. The main objective was to understand how these collaborations influenced digital consumer behavior, focusing on UGC creation and purchase intention. To this end, the adopted methodology was exploratory and descriptive, with the application of an online questionnaire to 493 respondents from the Southeast region of Brazil. The collected data were analyzed using descriptive statistics. The results indicated that, although “collabs” generated engagement among consumers, the production of active UGC was low, with passive interactions such as likes, saves, and quick reactions predominating. Additionally, collaborations showed limited influence on purchase intention and willingness to pay more for the products. It was concluded that, despite the strategic potential of “collabs” for the cosmetics market, their effectiveness depended on factors such as perceived relevance, brand identification, and the presence of additional stimuli for user content creation.

Keywords: Online behavior; Co-branding; Consumer engagement; Purchase intention; Digital interaction.

Sales and inventory planning proved relevant for business decision-making. The work aimed to develop a sales forecasting model by product, customer, and store for the subsequent 12 months, using historical monthly data. The methodology was based on applying models from the ARIMA family to 5,669 sales records, covering 745 unique product, customer, and store combinations, from January 2021 to December 2024. Exploratory analysis of the aggregated series, seasonal decomposition, stationarity tests, autocorrelation analysis, and identification of candidate models were performed. Model selection was supported by a stepwise procedure and walk-forward comparison with a uniform estimator (SARIMAX). The results indicated that the SARIMA(1,1,0)(1,0,0,12) model showed superior performance to ARIMA(1,1,1), with a MAPE of 13.51% and better adherence to seasonal behavior. In granular application, 122 series received SARIMA, 247 received ARIMA, and 376 used the historical average. It was concluded that the incorporation of the seasonal component strengthened the methodological consistency and practical utility of the model for sales forecasting.

Keywords: inventory planning; SARIMA; seasonality; time series.

The availability of consistent environmental data is fundamental for the advancement of research in climatology and water resource management. A computational algorithm in R language was developed and implemented for the automated acquisition, standardization, and organization of daily historical series of precipitation, level, and flow, from the WebService of the National Water and Basic Sanitation Agency (ANA). The method was based on a structured data pipeline, operating according to the time windows of 1961–1990, 1991–2020, and 2021–2024. The implementation used parallel processing, which ensured performance gains compared to sequential methods. As a result, the algorithm qualified the station inventory and generated structured data in Parquet format with gzip compression, ensuring high performance and volumetric economy. It was concluded that the automated solution overcame the operational limitations of manual and pre-existing tools, providing curated and high-granularity datasets, suitable for advanced hydrological modeling applications and Machine Learning algorithms.

Keywords: Automation; data science; Hydrology; R language; Time series.

Time management is configured as an essential element for improving productivity and the quality of deliverables, by allowing the organization, prioritization, and efficient execution of activities, directly contributing to the achievement of goals and the optimization of results in the professional and personal context. The objective was to understand how time management practices, tools, and techniques, adopted by rural professional training and social promotion instructors, were associated with two central outcomes: the completion of planned daily activities and the perception of the impact of these practices on well-being and productivity. Data were collected through a structured online questionnaire, composed of 18 questions, and analyses were performed in the Python environment with statistical libraries. The results revealed that high academic training and professional maturity did not translate into greater effectiveness in time management. Broad knowledge of tools and techniques was observed, but their practical use was limited and unequal. Simpler and more intuitive tools, such as to-do lists, showed greater adherence, while techniques requiring behavioral changes, such as the Pomodoro Technique, had low application. The use of digital solutions predominated, with emphasis on Google Calendar. It was evident that there was an almost unanimous perception that time management directly influences productivity and well-being. It was concluded that improving time management depended not exclusively on the adoption of tools and techniques, but, above all, on the development of practical and behavioral skills, combined with adjustments in working conditions.

Keywords: 5W2H; Eisenhower Matrix; GUT Matrix; Multitasking; Procrastination.

The need to optimize resources and understand complex criminal dynamics motivated the exploration of forensic data, often underutilized in textual reports. Multiple Correspondence Analysis (MCA) was applied to forensic attendances in the Bauru Macroregion, SP. The objective was to investigate and identify patterns of association between the type of occurrence, time of day, day of the week, and location, aiming to support operational decisions. Data from the Superintendence of Technical-Scientific Police (SPTC), collected between May 2023 and January 2026, were used. The methodology included data cleaning, transformation, and categorization of variables, as well as the construction of contingency matrices processed in Python. Initial descriptive analysis revealed that thefts and collisions predominated in the total volume of occurrences, while violent crimes, such as homicides, rapes, and femicides, represented a smaller portion but demanded significantly longer response times. MCA visually mapped the structural correlations, demonstrating clear associations between nighttime and weekend crimes with specific regions. Two main clusters were identified: a nighttime and weekend profile, associated with crashes, hit-and-runs, and suicides, with a focus on Bauru; and a daytime weekday profile, dominated by thefts and property damage. It was demonstrated that the use of exploratory machine learning techniques transformed raw forensic data into vital strategic information, providing support for efficient team allocation and underpinning the development of evidence-based public policies.

Keywords: Violent crimes; Contingency matrix; Service prioritization; Python; Response times.

The NBA’s Most Valuable Player (MVP) award holds significant relevance in professional basketball, reflecting athlete recognition and voting standards. A study developed and evaluated supervised machine learning models to predict the NBA MVP award, treating the problem as a ranking task based on the estimation of the voting “Share” value. Five algorithms were trained and evaluated – Random Forest, LightGBM, XGBoost, CatBoost, and NGBoost – using historical NBA data from 1996 to 2025, collected from the “Basketball Reference” portal. The models were tested on the 2019 to 2025 seasons. CatBoost achieved the best performance in Top-5 accuracy (77.14%), consistently identifying the top five candidates. Random Forest, LightGBM, and NGBoost tied in Top-1 accuracy (42.86%), correctly predicting the winner in three out of the seven seasons evaluated. The analysis of variable importance indicated that individual offensive performance, especially points per game, and team collective performance, represented by the win percentage, were the main determinants of the predictions. The results demonstrated the viability of the proposed approach and its high potential for future studies.

Keywords: Machine Learning; Boosting; NBA; Sports prediction; Random Forest.