A Lightweight Model for Detecting Cyberthreats Using Machine Learning Techniques

This research presents an innovative approach to enhance the security of Industrial Internet of Things (IIoT) networks through advanced intrusion detection. Utilizing the CICIDS2017 dataset, this study aims to apply machine learning classifiers such as Support vector machine, Decision Tree, and Random Forest algorithms, to accurately detect and classify various types of attacks like Brute Force FTP, Brute Force SSH, DoS, Heartbleed, Web Attack, Infiltration, Botnet and DDoS. The methodology used in this paper includes Data preprocessing which involves concatenating all CSV files, dropping redundant entries from the dataset and normalization using min-max normalization. Feature selection is applied on the dataset using Chi-Squared test and Principal Component Analysis. The challenge of class imbalance is overcome using the stratify parameter while splitting the data into train, test and validation sets in the ratio 60:20:20. The validation set enables the model to overcome overfitting thus achieving high accuracies of 94.3% with Support vector machine and 99.8% with Decision tree and random forest. This effort makes a substantial addition to the world of cybersecurity by demonstrating the efficacy of combining several analytical methodologies to improve IIoT security.