About

This is a comprehensive Olist Brazilian e-commerce analytics and churn prediction canvas that performs end-to-end data engineering, exploratory analysis, RFM segmentation, and machine learning modeling. The workflow downloads 9 CSV datasets from a public repository, conducts data quality profiling and relationship inference, builds a master table of 2017 delivered orders, calculates customer RFM metrics and churn labels (based on recency > 90 days), generates ~20 exploratory charts and visualizations (revenue trends, geographic distribution, category analysis, payment patterns, delivery delays), and trains two versions of predictive models: v1 (featuring recency, which reveals data leakage with perfect AUC ≈ 1.0) and v2 (leakage-free, using only behavioral features like frequency, monetary, and review scores, yielding honest AUC ≈ 0.56). The canvas documents findings through Markdown blocks, produces segment-level insights via RFM analysis (Champions, Loyal, At Risk, Lost), and outputs churn risk tables and ML metrics for downstream decision-making.