Multiclass Classification of Malware using Machine Learning

This work aims to classify malware into one of its nine families using multiclass classification using machine learning. Due to the rising number of malware assaults on numerous devices, including workstations, servers, cloud instances, and mobile devices, endpoint security is a critical concern for organizations today. Finding and stopping malware activity has become crucial to guaranteeing data security, privacy, and protection to defend against such attacks. It is even more critical to take safeguards now because of the wide variety of virus activities occurring across online and offline files. Using machine learning algorithms to forecast malware effectively by using malware detection datasets is one of the most efficient approaches to deal with this problem. Large amounts of data may be analysed by machine learning algorithms, which can also spot patterns that humans might find difficult to see. These methods enable more precise malware prediction and detection, which is crucial for endpoint security. Given that it is an affordable and effective option for endpoint security, the use of machine learning algorithms in malware detection is growing in popularity. Therefore, this work aims to employ three machine learning algorithms for analysing and classifying the malware present in a dataset. The dataset used in this work is the Microsoft BIG dataset which contains malware files belonging to nine different families. The malware files are converted to a feature dataset using feature selection and normalization techniques. This feature dataset is then exposed to classification algorithms like Random Forest, LightGBM and XGBoost which are known for their performance. Each model is assessed on the log loss of overall prediction. In this work, XGBoost showed the lowest log loss with value 0.046699789818985364.