APPLICATION OF ENSEMBLE LEARNING TECHNIQUES FOR PREDICTION MODELS

Abstract

With the exponential proliferation of malware, it has become a big concern in our everyday lives, which are largely reliant on computers running a variety of different types of software to function properly. Malware authors produce dangerous software by inventing new variations, new innovations, new infections, and more obfuscated malware through the use of tactics such as packaging and encrypting techniques, amongst other methods. Malicious software categorization and detection are critical components of cyber security research, and they represent a significant problem. Because of the rising number of false alarms, proper categorization and detection of malware has become a major issue that must be addressed in the near future. In this study, eight malware families were identified and classified according to their family members. The research presents four feature selection techniques for use in multiclass classification problems, each of which is designed to choose the best feature. Then the top 100 characteristics of these algorithms are picked for performance assessments and they are found. In order to determine the best models, five machine learning methods are compared. Then, using the feature ranking of the best model, the frequency distribution of features is determined. Finally, it is stated that the frequency distribution of each character in an API call sequence may be utilized to classify malware families.