Publication

Abstract : The increase in internet penetration has resulted in it becoming an indispensable part of everyday life. In 2022, more than 5.3 billion people counted on the internet- the ITU confirms that. While it offers various services over communication, e-banking, e-commerce, education, and healthcare-skimming from vulnerabilities become new grounds for malicious attackers to prey upon for either ransom or theft of sensitive details. Cyber threats are increasingly intricate in terms of form and conduct; hence, old detection systems are neither easily adaptable nor very precise. To this end, we put forth an ensemble-based intrusion detection system supported by deep neural networks to detect and counter more complex attack patterns. The ensemble method combining various classifiers allows for better generalization through leveraging multiple strengths of classifiers which provide versatility in terms of what attack vectors are captured. Oversampling methods are used to mitigate dataset imbalance while minimizing the bias against the minority class, and this is in tandem with anti-overfitting measures like Random Forest. We have used time-windowed feature aggregation during data preprocessing to help capture the temporal patterns characteristic of an APT, enabling more effective detection. A series of classic metrics like accuracy, precision, recall, F1-score, and confusion matrix measure the efficiency of a system. The conception of this framework relative to existing IDS solutions would certainly be one step ahead for the ones imposing AI in cybersecurity to prove their robustness and adaptability.