A NEW RESILIENT SMART MALWARE DETECTION USING DEEP LEARNING

Abstract

Now-a-days to observe cyber-attack are the usage of static and dynamic evaluation of request data. Static evaluation is primarily based on signature which we will suit current assault signature with new request packet information to become aware of packet is ordinary or consists of assault signature. Dynamic evaluation will use dynamic execution of application to notice malware/attack, however dynamic evaluation is time consuming. To overcome from this hassle and to enlarge detection accuracy with ancient and new malware attacks, we are the use of computing device mastering algorithms and evaluating prediction overall performance of a variety of computer gaining knowledge of algorithms such as Support Vector Machine (SVM), Random Forest, Decision Tree, Naïve Bayes, Logistic Regression, KNearest Neighbours and Deep Learning Algorithms such as Convolution Neural Networks (CNN) and LSTM (Long Short-Term Memory). Among those, quite a number fashions Deep studying CNN resulted in most suitable overall performance in contrast to different models. To put into effect this work and to consider computer getting to know algorithms overall performance this work the usage of binary malware dataset known as ‘MALIMG’. This dataset consists of 25 households of malware and utility will convert this binary dataset into grey pictures to generate educate and take a look at fashions for laptop gaining knowledge of algorithms. This algorithm changing binary statistics to photographs and then producing model, so they are referred to as as MalConv CNN and MalConv LSTM and any other algorithm refers as EMBER. Application convert dataset into binary pix and then used 80% dataset for education mannequin and 20% dataset for testing. Whenever we add new take a look at malware binary facts then utility will observe new take a look at statistics on teach mannequin to predict malware class.