URL-BASED HYBRID MACHINE LEARNING PHISHING DETECTION SYSTEM

Abstract

By using a large dataset based on the fishing url, they begin attacks on the Internet. This work is trying to detect cyber threats such as different types of machine learning methods, such as Decision Tree, Linear Regression, Random Forest, Naive Bayes, Gradient Boosting Classifier, Support Vector Classifier, and a new hybrid LSD model. As an extension, we have used a hybrid model that combines predictions of many individual models. This is achieved through rigid cross-fold validation and Grid Search Hyper parameter Optimization. Such a model makes classification stacking, which uses a dress technique to combine Random Forest Classifier and MLP Classifier, two base classifiers. As a meta-estimator, it appoints the LGBM classifier to reach the final prediction, which extends the project's ability to perform better classification. The effect of the model is evaluated using matrix including F1-score, recall, accuracy, and precision. The results show that the Hybrid LSD model effectively reduces the risk of fish attacks and provides strong protection against the ever -changing cyber danger. This study contributes to the development of better cyber security measures, and shows how you can improve the safety of the Internet by learning machine.