IntelliMetro‐Hybrid: A Machine Learning and Deep Learning Fusion Model for Economic Optimization in Smart Metro Systems

ABSTRACT Accurate anomaly detection in metro systems is crucial for ensuring operational safety, minimizing costly equipment failures, and enhancing predictive maintenance strategies. Despite the promise of existing machine learning (ML) and deep learning (DL) techniques, their effectiveness is often constrained by imbalanced datasets, temporal dependencies, and heterogeneous sensor data. To overcome these challenges, we propose IntelliMetro, a novel hybrid ensemble framework that seamlessly integrates tree‐based ML models with deep neural networks. IntelliMetro is rigorously evaluated against six classical ML models (XGBoost, Decision Tree, K‐Nearest Neighbors, Linear Regression, Support Vector Machine, Random Forest) and three DL architectures (ANN, LSTM, CNN) using the MetroPT‐3 dataset high‐resolution multivariate time series dataset capturing sensor readings from metro air compressors. The proposed IntelliMetro system consists of two main phases: the first phase involves the application of tree‐based models, such as Random Forest and XGBoost, to extract considerable patterns from the sensor data; and the second phase involves the combination of these features, followed by classification of anomalies with high accuracy using a light‐weight deep neural network. Experimental results demonstrate that IntelliMetro achieves state‐of‐the‐art performance with 98.7% accuracy, 98.3% precision, 99.3% recall, and 99.0% F1‐score, outperforming baseline models by 12%–18% in F1‐score. Notably, the framework reduces training time by 37% compared to pure DL models, while preserving interpretability through feature importance analysis. Its robustness is further validated under real‐world conditions, including sensor noise and temporal drifts. These findings underscore IntelliMetro's potential to revolutionize predictive maintenance in transit systems by reducing unplanned downtime (projected 22% cost savings) and enhancing passenger safety. This work advances ensemble learning for industrial IoT applications and provides a scalable template for anomaly detection in critical infrastructure systems.