Context-Aware Anomaly Detection in Smart Cities Using Multi-Modal Machine Learning Approaches

Abstract

Anomaly detection in smart cities is crucial for identifying unusual patterns in real-time data streams generated by diverse urban systems, such as traffic flow, energy consumption, air quality, and public safety. This study proposes a multi-modal machine learning framework for context-aware anomaly detection, integrating Convolutional Neural Networks (CNNs) for spatial feature extraction, Long Short-Term Memory (LSTM) networks for temporal pattern recognition, and contextual data (e.g., weather, public events) to improve detection accuracy. The hybrid CNN-LSTM model captures both spatial and temporal dependencies. At the same time, the inclusion of contextual information enables the model to adapt to changing conditions, improving the detection of anomalies such as traffic accidents or pollution spikes. Experimental results demonstrate that the proposed framework outperforms traditional anomaly detection methods in terms of accuracy, precision, and recall. The hybrid model's superior performance highlights its potential for real-time applications in smart cities, including sustainable urban management, fraud detection, and public safety monitoring.