The Internet plays a crucial role in our daily lives, prompting browsers to introduce features that may expose users to threats from malicious URLs. These links can lead to spam, malware, and phishing attacks, causing significant financial loss and personal data theft. Timely detection of such security threats is vital for protecting users and organizations. This research focuses on detecting malicious URLs through binary and multi-class classification using machine learning (ML) techniques, supported by exploratory data analysis and feature engineering. Various features, including lexical and network-based characteristics, are analyzed to improve URL classification. Additionally, the study employs a multivariate filter-based feature selection method to enhance accuracy by eliminating redundant features and validating them statistically. The combination of filter and wrapper methods aims to optimize feature selection for higher detection rates. Given the increasing risk of cyberattacks, especially with the rise of the Internet of Things (IoT), early detection methods are essential. Ultimately, this research develops and evaluates a multi-class classification model to identify web attacks such as phishing, spam, and malware from malicious URLs.