11/04/25 - Seminário do Programa Interinstitucional de Pós-Graduação em Estatística (PIPGEs) da UFSCar-USP

Nome do Palestrante: Anderson Ara (DEST-UFPR)

Título: Active Learning And Dimensionality Reduction: A Case In Process Mining

Resumo: This talk presents an integrated methodology that combines Active Learning and Process Mining to optimize efficiency and performance in CRM systems. Departing from traditional model-centric approaches, this study emphasizes the strategic use of t-SNE and entropy-based methods to enhance the GSx sampling step, allowing the selection of the most informative data points and reducing the dependency on large training datasets. The methodology is demonstrated through the transition of a Brazilian company from a telephony-based system to an omnichannel CRM system. By analyzing event logs, dialer activity, and call history, the study estimates success rates and identifies key process optimizations. The proposed approach leverages t-SNE for dimensionality reduction and entropy to capture uncertainty, ensuring a more robust sampling process during the GSx phase. Four predictive models Support Vector Machine (SVM), Random Forest (RF), Multilayer Perceptron (MLP), and Linear Regression (LR) were evaluated, with LR and SVM (linear kernel) demonstrating the most effectiveness in reducing unnecessary calls and enhancing dialer performance. Process mining further identified critical variables influencing success rates. This combined approach significantly minimizes the need for extensive datasets, improves model performance, and facilitates broader applications across CRM systems, streamlining the optimization of dialed calls without overloading computational resources. Joint work with Rafael Magalhães UFPR.

Todos(as) são bem-vindos(as)!