Research on traffic flow prediction based on the PLO-ICEEMDAN-KAN model

Accurate traffic flow prediction is crucial for improving the operational efficiency of intelligent transportation systems. However, the nonlinear characteristics and stochastic nature of traffic flow data make prediction highly challenging. This study, based on various advanced machine learning techniques, overcomes the limitations of traditional models in handling complex traffic data and constructs an efficient traffic flow prediction model: the PLO-ICEEMDAN-KAN model. The model employs the Polar Lights Optimization Algorithm (PLO) with the Minimum Envelope Entropy (MinEE) as the optimization objective. It optimizes both the noise level and the number of noise additions for the Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN). This optimization allows ICEEMDAN to decompose traffic flow sequences into multiple, more predictable sub-sequences based on optimal performance. After the decomposition is completed, the underlying Kolmogorov–Arnold Networks (KAN) model adaptively trains on these sub-sequences and generates prediction results. Finally, the proposed model reassembles the predictions from multiple sub-sequences to produce the final forecast outcome. Experimental results on real-world datasets demonstrate the superior predictive performance of the PLO-ICEEMDAN-KAN model, validating its potential application in intelligent traffic management.