It’s getting harder and harder to tell safe websites from possibly dangerous ones amid the thousands of new ones that are created every day. These websites frequently gather user data, and users’ private information may be hacked in the absence of sufficient cybersecurity safeguards including effective malware detection and classification. In order to improve cybersecurity, this work attempts to create model based on temporal convolutional neural networks for the effective detection and categorization of dangerous URLs. Within the perspective, the work combines support vector machine, random forest, decision tree, and k nearest neighbor in combination with the Bayesian optimization to effectively categorize URLs. Instance selection techniques are used to increase computational efficiency. The outcomes demonstrate how well random forests perform in term of high recall, precision, and F1 scores. Support vector machines, on the other hand, offer cheap performance at cost of longer training times. The findings further highlight the instance selection method's significant influence on these models’ performance, underscoring its importance in the machine learning pipeline for the classification of dangerous URLs.