Optimized Model Identification for Malware Detection in a File Using Machine Learning

Malware remains one of the most critical digital threats today due to the rapidly evolving landscape of online activities. Detecting malware has become a pivotal aspect of safeguarding computer system, especially as polymeric malware arises by utilizing dynamic modifications to confound traditional signature-based detection methods. To combat the surging volume of malware, Machine Learning (ML) has emerged as a key strategy. Hence, ML-based detection mechanisms are imperative. This overview underscores the diverse array of machine learning-driven detection techniques. The primary objective is to construct an optimal machine learning representation for malware detection within files. Three distinct machine learning algorithms are employed: random forest, logistic regression, and a basic neural network. From these, the superior model is determined based on rigorous evaluation metrics-accuracy, f1 score, and both normalized and non-normalized confusion matrices on both training and testing data. The selection process hinges on achieving the highest accuracy and f1 score. The random forest algorithm was depicted to be having highest train accuracy with 98.28% and test accuracy with 98.38% and f1 score was found to be 97.3% which was chosen as a final model for malware detection in a file. The chosen model subsequently serves as the foundation for malware detection.