Evaluating Machine Learning Algorithms with Feature Selection for Cybersecurity: A Case Study on Pegasus Spyware

The growing incidence of mobile spyware attacks presents serious threats to user privacy and data security in today's digital environment. It is crucial to identify such spyware, especially complex varieties like Pegasus, in order to safeguard private and corporate data. The goal of this work is to improve spyware detection by using and contrasting three machine learning methods to detect malware activity across several datasets: Random Forest (RF), Logistic Regression (LR), and K-Nearest Neighbors (KNN). A considerable amount of research identifies the difficulties in this field, such as poor accuracy, excessive false positives, and a lack of multi-model comparisons that make it difficult to reliably identify spyware in many data contexts. To address this, study provides a novel approach that compares these models using dataset of Pegasus spyware. This method uses feature selection and Cross-Validation for improved performance and interpretability, is among the first to offer a reliable multi-model reference for malware detection. According to the results, the RF model has the best detection ability in all circumstances, achieving the highest accuracy of 97.7<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> on Dataset 1, 85.6% on Dataset2, and 89.75% on Dataset3. Additionally effective are KNN and LR, which achieve accuracies of up to 89.39% and 89.72%, respectively. Our results show that RF has the potential to be a strong, dependable spyware detection solution, surpassing the 90% accuracy of previous models and establishing a new standard for future study and use in mobile security.