AI-Driven Cyber Threat Detection: Revolutionizing Security Frameworks in Management Information Systems
Mani Prabha, Md. Sazzad Hossain, Md Samiun, Mohammad Abu Saleh, Sweety Rani Dhar, Md. Abdullah Al Mahmud · 2024
Management Information Systems (MIS) in today's sophisticated cyber risk landscape are at risk, exposing the need for sophisticated and adjustable security solutions to meet these threats. The framework proposed in this paper works to increase the accuracy and efficiency of cybercrimes identified and mitigated through an AI-enhanced framework. This study explores and implements dimensionality reduction via Principal Component Analysis (PCA) for high dimensional data handling and Local Interpretable Model-agnostic Explanations (LIME) to increase model explainability using the CICIDS 2017 dataset. This approach enables cybersecurity professionals to understand the prediction because there is transparency, and it can trust the automated threat detection. Multiple machine learning models are evaluated, including XGBoost, Random Forest, Support Vector Machines (SVM), and K Nearest Neighbors (KNN). XGBoost achieved a near-perfect accuracy of 99.99% on these, and so might be able to classify these as they do cyber threats accurately. This proposed framework combines PCA and LIME in a new configuration specifically suited for real-time MIS applications: this makes it possible to achieve both high accuracy and interpretability in the face of various attack types using the CICIDS 2017 dataset. Considering interpretability, this analysis emphasizes cybersecurity in which transparent decision-making models allow professionals to understand, validate, and respond convincingly to detected anomalies. Combining robust interpretability tools with more advanced AI techniques can yield strengthened cybersecurity resilience in MIS and should be part of a client's toolbox. This framework integrates these methodologies to rapidly and accurately detect and manage real-time threats with explainability, thus improving MIS defenses against more sophisticated cyberattacks. This work provides the basis for future research, including model efficiency optimization and exploring other explainable AI techniques for broader cybersecurity applications.