Real-Time Anomaly Detection in IoT Sensor Data via Deep Learning and Deployment with Python Frameworks

The rapid proliferation of Internet of Things (IoT) devices has generated massive volumes of time-series sensor data that require continuous, intelligent monitoring. This project presents a comprehensive real-time anomaly detection system for IoT sensor data leveraging Long Short-Term Memory (LSTM) Autoencoder neural networks. The system processes multivariate sensor streams — including temperature, humidity, air quality, light intensity, and loudness — and identifies deviations from learned normal patterns using reconstruction error thresholding. The trained model is deployed through an interactive Streamlit-based web application, enabling users to upload CSV datasets, trigger analysis, and explore results through rich interactive visualizations without requiring programming knowledge. Experimental evaluation on a dataset of 6,558 time-series entries yielded an anomaly detection rate of 5.00% at a reconstruction error threshold of 0.002154, demonstrating high detection sensitivity with minimal false positives. The end-to-end pipeline — from raw sensor ingestion and MinMax normalization to sequence generation, LSTM inference, and downloadable labeled outputs — validates the practical viability of deep learning for IoT anomaly monitoring in smart environments.