ThreatScan-Strengthening Website Security with Predictive Analytics Using Machine Learning Model

ThreatScan is a machine learning-based approach proposed to strengthen website security by predicting potential external attacks. Motivated by the growing cyber threats and reliance on websites, ThreatScan utilizes the “ScrapingBot” at its core. This bot extracts malicious parameters from input websites, which are then fed into trained models such as decision trees, MLPs, and random forests based on their characteristics. A pipeline model is also employed, seamlessly integrating preprocessing with machine learning models like Logistic Regression, Decision Tree, and Random Forest for streamlined training. Then the model will predict potential threats generating lists of risks of each model as output. The Extensive evaluation demonstrates ThreatScan's effectiveness, with the decision tree model achieving 97.68% accuracy on the UNSW-NB15 dataset, the MLP model reaching 99.73% accuracy on KDD Cup 1999, the pipeline model attaining 97.55% accuracy on DDOS-Network-Logs, and the random forest model scoring 93.20% accuracy on the Malicious URL dataset. We‘ve also tested various parameters crawled by our ‘ScrapingBot’ from the nasa.gov website and have revealed significant insights. The source code and models are publicly released: ThreatScan GitHub Repository.