In order to improve the performance and interpretability of malware detection models, an intelligent fusion detection method based on data preprocessing and model optimization has been proposed. Multiple machine learning algorithms were integrated in this method, with key features being extracted using the random forest algorithm. Parameter tuning was performed with the ivy optimizer algorithm (IVYA) to construct a fusion model that enhances detection effectiveness. The SHAP(shapley additive explanations) method was utilized for interpretability analysis, thereby improving the transparency and reliability of the detection results. Experimental results show that accuracy, precision, recall, and other metrics exceed 98% across multiple datasets, demonstrating a significant advantage over traditional methods. Stronger protection and higher interpretability for cybersecurity are provided by this approach.