Assessing the Performance of Machine Learning Approaches for Cyber Attack Detection to Improve Cybersecurity

Cybercrime is a major problem all around the world since it causes countries and people to lose a lot of money every day. Recent developments have brought security models and prediction tools based on AI to address the difficulties of cyberattack detection and prevention. The purpose of this research is to assess how well a CNN model can identify cyberattacks on the CSE-CIC-IDS2018 dataset. The dataset underwent extensive preprocessing to address duplication, missing values, and feature normalisation using Min-Max scaling. An 80-20 train-test split was applied, and the CNN model was implemented alongside comparison models, including Deep Neural Networks (DNN), NB, and LSTM networks. Performance measures such as F1-score, recall, accuracy, and precision were used to assess the model’s efficiency. The CNN model had better results than its competitors in recognising and categorising cyberattacks, with a $95 \%$ F1 score, $99 \%$ recall, $92 \%$ precision, and 98.31% accuracy. Visualisation of the training and testing trends highlighted the model’s robustness with minimal overfitting, though some fluctuations in testing loss indicated areas for further optimisation. These results establish the CNN model as a reliable and robust approach for enhancing cybersecurity through efficient cyber-attack detection.