botxplo01/finwatch-zambia

A Machine Learning-Based Financial Distress Prediction System for Small and Medium Enterprises in Zambia # FinWatch Zambia > **ML-Based Financial Distress Prediction System for Zambian SMEs** --- ## Overview **FinWatch Zambia** is a production-deployed, full-stack machine learning system designed to predict financial distress in Small and Medium Enterprises (SMEs) within Zambia. It features a dual-portal architecture serving both business owners and institutional oversight bodies, combining classical financial ratio analysis with SHAP-based explainability and a multi-tier NLP narrative engine. The system is fully cross-platform, available as a professional web portal and a native Android application, featuring robust 30-day persistent sessions and a hardened environment-aware API. Developed as a Bachelor of Science in Computing (BSc BCOM) dissertation project at **Cavendish University Zambia**, 2026. --- ## Key Features - **Institutional Umbrella Architecture** - **SME Portal**: Company profile management, financial data submission, interpreted risk assessments, and prediction history with robust persistence. - **Regulator Portal**: Accessible via `/regulator`. Full systemic oversight, monthly distress trends, and anonymised anomaly flags (Emerald Theme). - **Policy Analyst Portal**: Accessible via `/analyst`. Read-only aggregate sector analytics and strategic reporting (Blue Theme). - **Native Mobile Experience**: Fully integrated with **Capacitor** for Android. Includes unclipped adaptive icons, native splash screen API integration, and mobile-optimized navigation. - **Robust Persistence Layer**: - **Persistent Sessions**: Mobile-only 30-day JWT sessions using dual-layer async storage (@capacitor/preferences + native file system). - **Prediction Persistence**: Retains manual financial inputs and extracted metrics across refreshes and navigations via `localStorage`. - **Explainable AI (XAI)**: Per-prediction SHAP attributions and global feature importance rankings. RANDOM_FOREST predictions take precedence on model disagreement. - **Environment-Awa …