Real-Time Energy Monitoring and Anomaly Detection in IoT-Based Systems : A Hybrid Approach Combining Isolation Forest and Dynamic Baselines for Sustainable Energy Awareness

Sammanfattning

With the growing global demand for energy, optimizing consumption has become a key challenge for individuals, businesses, and policymakers. Inefficient usage not only increases costs but also negatively impacts the environment. This thesis presents the design and development of a real-time energy monitoring and anomaly detection system built on Internet of Things (IoT) infrastructure. It aims to enhance energy awareness and support proactive decision-making by providing live insights into power consumption and alerting users to abnormal usage patterns. The prototype uses smart energy monitoring devices to collect real-time power data, transmitted via MQTT to a centralized backend. Data is stored in InfluxDB and visualized through an interactive web dashboard. A REST API supports user and device management, enabling multi-device, multi-user scalability. Users can monitor consumption trends and receive real-time anomaly alerts for faster response. A hybrid anomaly detection approach combines an AI-based Isolation Forest model with dynamic baseline thresholding. The Isolation Forest identifies outliers using patterns in historical data, while the baseline method detects deviations via a rolling median. Anomalies trigger real-time alerts delivered through WebSocket notifications. The project follows a Design Science Research methodology, emphasizing iterative development and evaluation. The system was validated using real-world data, demonstrating effective anomaly detection, responsive performance, and user-friendly visualization. This thesis shows that combining lightweight statistical methods with AI-driven techniques in a real-time IoT framework can promote energy-efficient behavior and support sustainability. The solution also offers a scalable foundation for future features such as anomaly severity levels, usage recommendations, and integration with smart environments.