| Sumario: | Objectives: This study aims to develop an Automated Traffic Prediction and Classification (ATPC) system that leverages advanced feature extraction, optimization, and classification techniques to address the related challenges. Methods: This framework takes from the Lebanon traffic and the road traffic severity prediction datasets for training and evaluation. First, a set of features is extracted from the data using the Behavioral Time Scale (BTS) feature extraction model, which identifies traffic occurrence. This algorithm improves prediction accuracy via evaluating the Kendall rank correlation measure, homogeneity, and time-series correlation values. Second, the Monkey Tree Selection Optimization (MTSO) algorithm, which estimates the best fitness value based on the position update of global "monkeys," is used to select the optimal subset of extracted features. Third, these optimal features are used to train the classifier, which reduces processing time while preserving--or even enhancing--prediction accuracy. Finally, the Propagated Deep Recurrent Network (PDRN) model is trained using the selected features to classify samples into 'normal' and 'traffic' categories. Results: Evaluations demonstrate that the proposed system achieves an accuracy of 98.86%, similarity coefficients of up to 98.9%, and an error rate as low as 2.5%, indicating significant improvements in both performance and efficiency. When compared to models such as Decision Tree, Logistic Regression, Random Forest, and XGBoost, the proposed MTSO-PDRN technique demonstrates higher accuracy, kappa coefficient, and specificity, along with a lower error rate. Furthermore, the PDRN method proved highly precise in classifying injury types. Conclusion: The enhanced performance and efficiency of the proposed system are promising, indicating its strong potential for predicting and classifying road traffic conditions to help prevent accidents.
|
|---|