Unveiling Malicious Intent: A Novel Approach using Artificial Neural Networks

Abstract

This study explores the use of an Artificial Neural Network (ANN) framework for malware detection. The dataset was preprocessed using a Label Encoder to convert categorical variables into numerical representations. It was obtained via Kaggle and included a mixture of malware and benign samples. Because the majority of machine learning algorithms, including ANNs, depend on numerical inputs to function properly, this conversion is crucial. Rectified Linear Unit (ReLU) activation function, which is well-known for being effective in training deep networks by reducing the vanishing gradient issue, is used in the input layer of the suggested ANN design. ReLU Is also used in a dense hidden layer, which gives the model the ability to recognize intricate correlations and patterns in the data.The last dense layer produces probabilities that can be understood as either benign or malicious classifications using a sigmoid activation function. RMSprop is the preferred optimizer because it dynamically modifies the learning rate during training to produce more reliable and rapid convergence. Binary cross-entropy is a useful loss function for binary classification problems because it incentivizes the model to generate confident and accurate predictions by penalizing inaccurate predictions proportionate to their confidence. The suggested methodology aims to solve the difficulty of identifying sophisticated and ever-evolving malware threats by accurately classifying instances as either benign or malware. In an ever-changing cybersecurity market, this technique seeks to provide an efficient malware detection solution by utilizing ANN’s flexibility and ability to learn from complex data.