Abstract: The traditional malware detection methods based on API (application programming interface) call sequences fail to sufficiently capture the long-term temporal dependencies and neglect the high-order nonlinear relationships among features. To address these issues, this paper proposes a multi-classification model for malware (TCN-SE-ChebyKAN) that integrates a temporal convolutional network (TCN) and Chebyshev-Kolmogorov-Arnold network (ChebyKAN). First, the TCN module is employed to extract features from API call sequences. By leveraging causal convolutions and dilated convolutions, the model captures temporal characteristics and long-range dependencies representing malware behavior, thereby obtaining more comprehensive behavioral information. Next, a squeeze-and-excitation (SE) network module is introduced to construct a channel attention mechanism. Through dynamic adjustment of channel weights, the model enhances its ability to capture discriminative features. Finally, the KAN (Kolmogorov-Arnold network) module is utilized to model complex relationships among features. By improving the KAN module with Chebyshev polynomials, the model strengthens its capability to model high-order nonlinear relationships among features, boosting overall detection performance. Experimental results demonstrate that the proposed model achieves an AUC value of 92.53% on the Mal-API-2019 data set, with significant improvements in other detection metrics.