Artificial Intelligence-Driven Neural Computing for Navigation Decision Classification in Autonomous Vehicles

Autonomous vehicles (AVs) require robust decision-making capabilities to navigate dynamic environments securely and effectively. Furthermore, advanced sensors and computational models are utilized to perceive the environment and predict the behavior of other road users. The main challenges include dealing with unpredictable factors such as weather fluctuations, traffic patterns, and road obstacles. While these technologies provide contextual awareness, existing techniques often face difficulties in handling complex sensor data under rapidly changing conditions and selecting the most useful features for navigation decision classification. To overcome these challenges, there is a need for reliable and intelligent architectures for navigation decision classification systems to support AVs by utilizing AI-driven machine learning (ML) and deep learning (DL) techniques. Currently, DL methods have gained significant importance in enhancing AV capabilities by enabling improved perception and decision classification from complex data. Therefore, this paper introduces an Intelligent Bidirectional Temporal Attention Framework for Navigation Decision Classification in Autonomous Vehicles (IBTAF-NDCAV). The major aim of this research is to design an effective navigation decision classification system for classifying driving actions using sensor data. To accomplish this aim, the proposed method applies improved mutual information-based feature selection to select informative features while reducing redundant ones. For classification, the proposed IBTAF-NDCAV method utilizes a bidirectional temporal attention convolutional network as an AI-driven neural computing model that captures temporal dependencies and effectively classifies navigation decisions. To enhance convergence and training stability, the Adamax optimizer is used to optimize the neural network training process. The experimental result analysis of the IBTAF-NDCAV technique is conducted on a benchmark Autonomous Driving Sensor-Kinematics dataset, which includes 5000 samples distributed across five classes: follow, yield, stop, lane change, and overtake. The comparative results demonstrate superior performance, achieving an accuracy of 96.76%, with class-wise precision, recall, F1-score values and MCC of 90.97%, 81.68, 85.31%, and 83.40% on the 20% testing set. The temporal attention mechanism highlights informative temporal patterns contributing to the navigation decisions, thus providing a basis for interpreting the model's decision-making procedure. Therefore, the result highlights the effectiveness of the model, making it appropriate for utilization in autonomous vehicle navigation and path-planning applications.