Published Paper Details:

This research presents a fully implemented, end-to-end smart energy monitoring system integrating machine learning, a real-time backend service, and a Flutter-based mobile application. The system utilizes the publicly available COMBED dataset to simulate real-time smart meter readings at 30-second intervals via WebSocket streaming. Four machine learning models--KMeans clustering, Isolation Forest anomaly detection, LSTM short-term forecasting, and ARIMA long-term forecasting--enable comprehensive energy analysis including pattern recognition, anomaly alerts, and multi-horizon predictions. A FastAPI backend provides REST endpoints and WebSocket support with model inference latency below 100 ms. The Flutter application delivers a mobile-first dashboard with real-time charts, anomaly lists, bill estimation in INR, and forecast visualizations. Twilio SMS integration provides automated alerts for critical anomalies. This work distinguishes itself from prior literature through complete implementation, real-time simulation using open datasets, mobile-first architecture, and deployable multi-model integration. The system is reproducible, scalable, and extendable to real IoT smart meter hardware.