International Journal of Science, Engineering and TechnologyAuthors: Teja B, Susmita Das, Rishabh Arya, Ujjwal Kumar
Abstract: Networked infrastructure has grown quickly, and along with it the scale and complexity of cyber-attacks, to the point where manually watching network traffic is no longer a workable way to catch threats in time. This paper describes a machine-learning-driven system for cyber threat detection that examines network flow attributes and sorts traffic into three categories: Normal, Distributed Denial-of-Service (DDoS), and Ransomware. Building the classifier involves a supervised pipeline — cleaning the data, encoding categorical fields, scaling features, comparing Random Forest, XGBoost, and Support Vector Machine (SVM) models, and finally serializing the winning model for deployment. Once selected, the classifier is wrapped in a Flask REST API and packaged with Docker, and a React.js interface lets a user enter traffic features and immediately see the predicted attack type along with its class probabilities. A review of recent 2026 literature points to ensemble methods, techniques for handling class imbalance, explainability, and adaptive detection as the areas currently shaping network-security research. With that context in mind, this work focuses on building a complete, deployable pipeline that links model training directly to a usable web-based prediction tool. On a dataset matched to the target schema, the gradient-boosted XGBoost model reached the best test accuracy at 98.62%, with Random Forest close behind at 98.46% and SVM at 96.83%.
