IMPLEMENTATION OF DEEP LEARNING METHOD FOR MALWARE DETECTION USING CNN ARCHITECTURE WITH HYPERPARAMETER TUNING

• Main Author: Hakim Khairul, Halimul
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/67891
• Institution: Institut Teknologi Bandung

Malware is computer software that can wreak havoc on an OS. One of the major security and privacy dangers to people, businesses, governments, and the larger community at the moment is malicious software, or malware. The AV-Test Institute has registered more than 350,000 malicious software programs and undesirable applications daily, according to data from 2021. According to the data, there will be an annual increase in the quantity of malware discovered. Advanced detection techniques must be used to offset the malware's quick evolution. One of the newest malware detection techniques makes use of machine learning techniques. It is asserted that this technology outperforms conventional signature-based detection methods. The malware samples or datasets required to train the model to identify malware are necessary in practice for the machine learning malware detection method. The final project of implementing a deep learning method for malware detection using CNN architecture with hyperparameter tuning is a subsystem that is used to evaluate malware image datasets. using a dataset consisting of 2 classes, namely malicious and benign. This study aims to analyze the type of image and parameters that affect the performance of the model before deployment. The analysis is carried out by comparing the evaluation results of the effect of choosing the type of image, the image resizing parameters, and the number of epochs. Based on the test results, the evaluation results of the best model with an accuracy metric of 90% can be achieved by selecting an RGB image and then applying the preprocessing parameter resizing the image to 128×128 pixels, where the number of epochs is 50. The model with the best evaluation is stored, and the model deployment is carried out. The deployment results show quite good results in terms of appearance and functionality of each component.