KANPoolConvNeXt: an innovative malware detection method

Abstract

The major objective of this research is to detect malware using an innovative deep learning model. Therefore, we present a new convolutional neural network (CNN) based on the Kolmogorov-Arnold Network (KAN). To the best of our knowledge, we are the first team to test a KAN-based CNN for malware detection. In this research, a publicly available malware dataset was used, and this dataset is in image format. Three cases were created by deploying this dataset. To compute classification results for these cases, a KAN-based CNN was designed. This deep learning architecture has four phases: (i) input, (ii) main, (iii) downsampling, and (iv) output. In the stem phase, the first tensor is created. The core innovation of this model is the main block, where two activation functions are utilized to generate feature maps. Downsampling halves the tensor width and height and increases its depth. The final phase of the proposed CNN is the output phase, where the classification results are generated. We applied this CNN to the malware dataset; therefore, this CNN is termed KANPoolConvNeXt. Using the dataset, three cases were created, and the proposed KANPoolConvNeXt attained over 97% classification accuracy in all three cases. The obtained classification results demonstrate that the proposed KANPoolConvNeXt is an innovative information security solution. This deep learning model can also be applied to other classification problems.