About

This analysis examines 99,441 Brazilian e-commerce orders to identify the drivers of negative customer reviews and build a predictive model for bad reviews (score ≤2★). Our findings reveal that delivery timeliness is the dominant factor: late deliveries increase the bad review rate from 9.2% to 78.4%, and the presence of customer comments strongly signals dissatisfaction. Using a Random Forest classifier trained on 76,665 orders, we achieve 84% accuracy and 83.8% ROC-AUC on held-out test data, enabling early identification of at-risk orders before delivery and flagging 61% of bad reviews with 42.6% precision—sufficient for operational triage and proactive customer intervention.