KNOXRT: Knowledge-Networked Optimization for eXploratory Real-time Training

Effective threat detection and classification are essential in real-time data stream processing. In this paper, KNOXRT, an architecture that combines traditional tree-based models, such as Random Forest, XGBoost, and CART, with deep learning models, including Convolutional Neural Networks (CNNs), Long ShortTerm Memory (LSTM), and Gated Recurrent Units (GRUs), is presented. Real-time data is ingested using Apache Kafka, and feature extraction is performed in parallel through CNN, CNN+LSTM, and CNN+GRU models. The resulting features are merged and passed through multiple classifiers, achieving high classification accuracy. The best-performing model, CART, achieved perfect training accuracy (1.00) and strong production testing accuracy (0.9927), precision (0.9927), recall (0.9927), and F1-score (0.9927). Class imbalances are addressed using GSmote for balanced training data, and real-time predictions are supported by containerized microservices. Continuous performance monitoring and online retraining ensure the system adapts to dynamic environments without interruption.