Efficient cluster-based deep anomaly detection based traffic analysis and multi-objective optimization for smarter traffic control

In the need for an efficient traffic management system, this paper introduces the Integrated Traffic Analysis and Optimization (ITAO) methodology. This innovative method combines two powerful techniques: Cluster-based Deep Anomaly Detection (CDAD) and the Nelder–Mead Pareto African Vultures Optimization Algorithm (NelPareto AVOA). Together, they tackle the major challenges of urban traffic. Using large amounts of traffic data from various sources, ITAO can identify complex traffic patterns, quickly detect unusual conditions, and make real-time adjustments. This paper details the steps of collecting and processing data, using CDAD for deep analysis and feature extraction. It highlights the use of a time-decay function in DBSCAN clustering and Temporal Deep Belief Networks to better understand traffic patterns and detect anomalies. Furthermore, ITAO employs a multi-objective optimization strategy to achieve several goals: maximizing traffic flow, reducing congestion, improving energy efficiency, and enhancing pedestrian safety. The NelPareto AVOA, inspired by nature, ensures that traffic signals and patterns are analyzed and optimized in real-time, adapting to the changing urban environment. Across various datasets, ITAO achieves superior predictive accuracy, consistently delivering the lowest Root Mean Square Error (RMSE) values. For instance, in PEMS-BAY, the RMSE is approximately 15.5 for a batch size of 10 and rises to 16.0 for larger batches. Similarly, in METR-LA, the RMSE starts at 14.0 for a batch size of 10 and increases to 15.0 at batch size 40. INRIX-SEA follows a similar trend, with an RMSE of 15.0 at a batch size of 10, reaching 16.0 for larger sizes. The RMSE for PeMSD4 begins at 14.5 and rises to 15.5, while for PeMSD8, it starts at 15.0, increasing to 16.0 for larger batches. In addition to its accuracy, ITAO excels in classification tasks, as illustrated by True Positive Rate (TPR) values. At specific False Positive Rate (FPR) levels, ITAO outperforms other methods with TPR ranging from 0.9 to 1.0, while methods like WKNN-FDCN achieve TPR values from 0.7 to 1.0. Other models perform less robustly: STAWnet shows TPR between 0.4 and 0.9, PGCN from 0.3 to 0.8, VDGCNet from 0.2 to 0.7, DSTF from 0.1 to 0.5, and DL from 0.0 to 0.4. These results, combined with ITAO's ability to optimize multiple objectives such as traffic flow, energy efficiency, and pedestrian safety, highlight its effectiveness over existing methods.