Séries temporais hierárquicas na previsão de denúncias de violência contra a mulher : estudo de caso com dados do serviço Ligue 180

Abstract

This work aims to study reports of violence against women in Brazil from January 2014 to March 2024. The primary objective is to analyze and forecast trends in these reports using hierarchical time series techniques and the ARIMA forecasting model at various hierarchical levels, including national, regional, and state. The research is applied and quantitative, based on data extracted from the Ligue 180 service. The methodology involves the implementation of approaches such as Top-Down, Bottom-Up, Middle-Out, and the Minimum Trace Optimal Reconciliation. These approaches enable a detailed comparison of different forecasting strategies, identifying the most effective for the context of violence reports. Descriptive data analysis reveals a continuous upward trend in reports, reflecting both increased awareness of violence against women and greater trust in the reporting system. The results indicate that the Middle-Out approach performed best, showing the lowest MASE and RMSSE values, suggesting better adaptability to the specifics of the data. Additionally, the research discusses seasonal and regional variations in reports, considering factors such as the impact of the COVID-19 pandemic and awareness campaigns.