A Hybrid Machine Learning Framework for Predictive Analytics in Large-Scale Information Systems

In today’s world of big data, massive information systems are being targeted by cyber threats which are becoming more and more sophisticated and, therefore, security and operational reliability have become a great challenge. A single, model predictive analytics approach cannot usually detect the complex and non, linear patterns that reveal the nature of heterogeneous network traffic. In this research paper, a new Hybrid Machine Learning Framework (HMLF) is suggested, which combines several base learners such as Decision Trees, Random Forests, Gradient Boosting, and K, Nearest Neighbors using ensemble methods like soft voting and meta-stacking. By using a thoroughly balanced UNSW, NB15 dataset, which represents various attack vectors in the modern world, the framework applies a multi, stage feature selection and standardization preprocessing pipeline. The experiments confirm that the combination of the proposed methods significantly exceeds the performance of the individual models, with a record of 98.42% accuracy and a high F1-score, while still being time, efficient enough for on, line operations. Moreover, the strength of the suggested model is assured through extensive cross, validation and statistical hypothesis testing, thus it can be regarded as an ultra, modern solution to the problem of predictive analytics in large, scale information environments.