Malware Detection by Machine Learning based on the LightGBM Model

Wannacry has grown to be considered one of the biggest threats affecting computer security over the past ten years, which has prompted continuous research on detection and mitigation techniques. The EMBER 2018 dataset, which contains sample WPE files categorized as harmless and harmful, is used to synthesize research findings and data in order to build an efficient machine learning model for malware identification. The pre-processed data is sent into a deep learning system for training. The machine learning model was built using dense and conventional dropout layers to reduce constraints on resources and generate quicker, accurate results. By evaluating the underlying API functionality that a file utilizes, we can determine if its behavior could be considered benign or malicious. LightGBM has been built utilizing over 1.55 million tagged example sets, each represented by 1,000 API-call-based attributes. Every model, including the Random Forest model, which displayed outstanding results with 99.49% reliability and fine-grained characteristics like precision and recall to lower mistakes, is assessed using conventional criteria for performance. The study idea is based on the implementation of contemporary standard machine learning models (such LightGBM) to guaranty the highest-level using malware identification accuracy and the usage of reputable software programs to extract standardized features from PE data files. The research also explores classification thresholds and performance optimization methods, aiming to provide a reliable and effective initial solution to the problem of static malware detection.