These days, one of the most well-known cyber threats is malware. When the amount of data grows, so does the number of malware threats. Malware not only increases in quantities but also becomes smarter and more difficult to detect. Detect malware threats on websites caused by high data traffic, becomes a challenging problem, which must be solved. Moreover, billions of dollars are lost annually due to malicious website scams. Applying analytics to discover new information, predict future malware insights, and make control decisions is a critical process that makes online websites secure. In this research, we propose and analyze a machine learning-based system to detect malicious website behavior based on specific features. With these features, websites are categorized as harmful or non-malicious. This paper employs several machine learning techniques, including Logistic Regression, Random Forest, and integration of both Random Forest and Logistic Regression Algorithms to classify malicious and non-malicious websites, based on various weight ratio selection circumstances to improve results. Applying cascading algorithms in a new way depends on that the first algorithm's prediction fed as input to the next algorithm. So, reasonable results are reached with 100% accuracy, recall, and precision and 0% False Negative Rate.