Author : Noman Abid

Network anomaly detection systems have grown in popularity and usefulness as a means to identify assaults, intrusions, and abnormalities in the ever-increasing volume of data sent by the internet and smart devices. Unusual patterns in network traffic may be reasonably anticipated using machine learning techniques. Nevertheless, the majority of prior efforts have been on anomaly detection inside conventional ML frameworks. Focussing on the IoT-23 dataset, which contains both harmless and malicious network traffic, this research explores the use of ML techniques for identifying and categorising anomalies in network security. Data preparation, feature engineering, model execution, and assessment are all a part of the technique. Models such as CNN, DT, LR, and SVM are employed to classify network anomalies, with performance evaluated using accuracy, precision, recall, and F1-score. Achieving a balance across parameters including F1-score, recall, and precision, the CNN model surpasses other models with an accuracy of 98.69%. Comparing the results, it can be concluded that CNN performs well aimed at anomaly detection, whereas Decision Trees offer a high level, and Logistic Regression and SVM are less accurate and stable. Since CNN is applied in deep learning, the study shows the effectiveness of deep learning models for network security in terms of anomaly detection with a suggestion to approach model optimisation to avoid overfitting and enhance generalisation.

Keywords: Cybersecurity, IoT, Machine Learning, Network Security, Malicious Activities, Anomaly detection, Classification.