Freshgeek52/sawa-citi-ml-framework

Machine learning framework for analyzing e-commerce customer behavior — Case study: Sawa Citi Supermarket, Rwanda # A Machine Learning Framework for Analyzing E-Commerce Customer Behavior ### A Case Study of Sawa Citi Supermarket, Rwanda > Master's Thesis in Computer Science — 2026 --- ## Overview This repository contains the complete thesis project: written chapters (Markdown source) and the Python machine learning pipeline that produces all empirical results for Chapters 4 and 5. **Case study:** Sawa Citi Supermarket (Brood Rwanda Limited), Kigali, Rwanda **Data:** POS transaction data from two branches — HQ (March 2026) and Kisimenti (March–May 2026) **Total records:** 26,533 line items across 16,597 transactions from 2,056 named customers --- ## ML Pipeline Results (Sawa Citi data) | Track | Algorithm | Key Metric | |---|---|---| | A — Customer Segmentation | K-Means (k=3) + RFM | Silhouette = 0.496 | | B — Purchase-Intent Classification | XGBoost (best) | ROC-AUC = 0.9995, F1 = 0.985 | | B — Purchase-Intent Classification | Random Forest | ROC-AUC = 0.9988, F1 = 0.966 | | B — Purchase-Intent Classification | Logistic Regression (baseline) | ROC-AUC = 0.964, F1 = 0.817 | | C — Product Recommendation | Item-Based CF (cosine similarity) | Precision@10 = 0.026, MAP = 0.062 | **Customer segments discovered:** | Segment | Count | Avg Monetary (RWF) | Avg Frequency | |---|---|---|---| | Champions | 363 | 101,548 | 4.3 days | | Potential Loyalists | 80 | 15,609 | 1.6 days | | At-Risk Customers | 1,613 | 5,298 | 1.1 days | --- ## Repository Structure ``` thesis/ ├── overview.md # Project rules ├── README.md ├── .gitignore ├── chapters/ # Thesis writing (Markdown source) │ ├── 01_introduction.md │ ├── 02_literature_review.md │ ├── 03_methodology.md │ ├── 04_results.md # pending — uses pipeline outputs │ └── 05_conclusion.md # pending ├── references/ │ └── references.bib # IEEE bibliography (single, end of document) ├── figures/ # Diagrams for the the …