Traffic accident prediction method based on improved support vector machine

To further enhance the accuracy and efficiency of traffic accident prediction, an improved support vector machine (SVM)-based approach is introduced. This method utilizes an encoding-decoding architecture to capture temporal data characteristics. Additionally, a spatial domain-based graph convolution operator is devised to address the issue of limited spatial connectivity between nodes, thereby extracting spatial features and furnishing reliable inputs for subsequent predictions. Leveraging the nonlinear nature of SVM models, a tailored traffic accident prediction model is constructed. To guarantee prediction precision, the ant lion algorithm is employed to refine the crucial parameters of the SVM model, bolstering its performance. The acquired time series prediction outcomes and spatial feature extractions are integrated to accomplish traffic accident prediction. The results show that the prediction time error result of the proposed method can always maintain below 0.8, the prediction position error result is less than ± 80cm, and the average prediction time is only 1.837s, which shows that the proposed method has high accuracy and efficiency of traffic accident prediction, and can provide strong support for traffic management and planning.