ADM District-Based Electricity Fault Prediction System

2026·Solo developer·Data ScienceEge Career Fair 2026 Hackathon

TechnologiesPython · Pandas · NumPy · LightGBM · CatBoost · Scikit-learn · Streamlit · Plotly · REST API

Problem

For electricity distribution companies, foreseeing faults and outages is critical to sending maintenance crews to the right place. The goal was to predict 7-day fault risk for 49 districts from historical outage and weather data.

Approach

I cleaned roughly 367,000 outage records from 2021–2024 with Python (Pandas, NumPy) and turned them into a complete time series. Through Open-Meteo API integration I collected historical and current weather data, and designed a 28-feature feature-engineering pipeline including lag, rolling-window, and seasonal variables. I developed a 7-day fault-risk prediction with a LightGBM + CatBoost ensemble that combines classification and regression.

Result

89%AUC-ROC

74%F1 score

49districts · ~367,000 records

I built a multi-page, interactive web dashboard with Streamlit and Plotly featuring a risk map, district-level detail analysis, and model-performance visualizations.

ADM fault prediction dashboard: district-level risk map, risk summary cards, and the list of high-risk districts — The fault prediction module: 7-day risk map, province-level summary, and the highest-risk districts.