Malware detection and classification have become increasingly complex due to the evolving strategies of malicious software, especially those targeting mobile platforms like Android. Traditional antivirus solutions are no longer sufficient, as the sheer scale and sophistication of malware attacks have outpaced their capabilities. To address this, the study presents a novel hybrid feature selection model – FilWrap, for robust and accurate malware classification using Android network traffic data. The FilWrap model combines the Chi-square filter method and Recursive Feature Elimination (RFE) wrapper method, further optimized through a Genetic Algorithm (GA). This approach improves both computational efficiency and predictive performance. The dataset used comprises 15,036 samples, including 5,560 malware and 9,476 benign applications, with 215 features extracted from network traffic. After data cleaning, feature scaling, and label encoding, feature selection was conducted in three stages, reducing the dataset to an optimal subset of 20 highly relevant features. Multiple classifiers, including K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), and Neural Network (NN), were trained and evaluated using these selected features. The FilWrap-based KNN model achieved a perfect classification performance with 100% accuracy and ROC-AUC. Other classifiers also demonstrated strong results, with Random Forest and Neural Network achieving 96.39% and 96.30% accuracy, respectively. Comparative analysis with benchmarked feature selection techniques confirmed that the proposed FilWrap model outperformed existing methods in terms of accuracy, feature relevance, and detection capability. These results validate the efficacy of integrating filter, wrapper, and evolutionary algorithms into a unified feature selection pipeline for enhanced Android malware detection.