AI-Driven Predictive Cyber Defense Framework Integrating Hybrid Learning Models for Real-Time Intrusion Analysis

Abstract

As cyber assaults are rising in complexity and number, there is a need for sophisticated proactive security measures that can effectively identify and predict the network attacks. The paper suggests the advanced AI based cyber attack prediction system built with CICIDS2017 dataset and based on ML, DL, Generative AI and Explainable AI. The entries are cleaned for missing values and duplicated ones and then labelled and normalized, and Principal Component Analysis is applied to the data to reduce its dimensionality for efficient learning. Different classifiers that rely on ML are applied like Decision Tree, Random Forest, Extra Trees Classifier, Logistic Regression, Gaussian Naive Bayes and a hybrid Voting Classifier (Random Forest, LightGBM and XGBoost). Furthermore, DL models are also employed including deep neural network (DNN), convolutional neural network (CNN), long-short term memory (LSTM), CNN-LSTM and CNNLSTM-GRU. We employ generative models to generate attack patterns (Variational Autoencoder, Generative Adversarial Network, DistilGPT2) and enhance anomaly representation. Experimental testing indicates that the Voting Classifier achieves 99.6% accuracy, whereas the LSTM model achieves 99.3% accuracy, a good detection rate in all attack categories, such as DoS, DDoS, PortScan, Bot and Infiltration. LIME and SHAP are used to guaranty the model interpretability. It has been implemented in Flask that enables authentication, processing of input, visualization and categorization of network traffic as benign or dangerous. “Keywords— Cyber attack prediction, ML, DL, Generative AI, Explainable AI, LSTM, Ensemble Learning, Intrusion Detection.”