Erhahi/Jumia-Jitters-Return-Prediction

A data science project for the DataVerse Africa Challenge that predicts product returns on Jumia, analyzes seller behavior, and provides actionable business recommendations using Python and machine learning. ___ # Dataverse Africa Challenge – E-Commerce Returns & Seller Profiling ## Overview This project analyzes real-world e-commerce data from Jumia to uncover return risk patterns and seller behaviors. It combines data analysis, predictive modeling, and business recommendations to help reduce logistics costs and identify risky sellers. ___ ## Phases Completed 1. **Data Exploration** – Explored 987 records from the Jumia dataset (`jumia_jitters_dataset.csv`) 2. **Market Insights** – Analyzed product prices, delays, customer reviews, and seller activity 3. **Modeling Return Risk** – Built logistic regression to predict return likelihood (AUC: 0.91) 4. **Strategic Recommendations** – Proposed actionable insights for seller risk monitoring 5. **Script Pipeline** – Packaged reusable Python scripts for scalable deployment ## Key Insights - **Late delivery** and **low customer ratings** drive return risk - Certain sellers have unusually high return rates - Sentiment analysis of customer reviews adds strong predictive power ## Files & Structure ___ ``` jumia-jitters-return-prediction/ ├── data/ │ └── jumia_jitters_dataset.csv # Raw dataset ├── cleaned/ │ ├── cleaned_orders.csv # Cleaned dataset │ └── seller_summary.csv │ └── seller_risk_action.csv │ └── risky_sellers.csv │ └── orders_with_return_risk.csv │ └── model_performance_metrics.csv │ ├── models/ │ └── return_prediction_model.pkl # Saved model │ └── scaler.pkl # Scaler object │ ├── scripts/ │ ├── config.py # Config variables │ ├── utils.py # Feature engineering & preprocessing │ └── run_pipeline.py # Script to run predictions │ ├── charts/ │ ├── top_states_bar_chart.png │ ├── avg_rating_by_state.png │ ├── top_returning_sellers.png │ ├── roc_curve_plot.png │ └── feature_importance.png │ ├── notebooks/ │ └── Phase_1_Data_Cleaning_Feature_Engineering_Er …