Malicious URLs have become serious threats to cybersecurity, also forming incubators for Internet criminal activities. With visiting malicious URLs, visitors may undergo illegal actions such as spamming, phishing and drive-by downloads which seriously threat visitors' privacy and security that cause losses of billions of dollars every year. Traditional methods such as using URL blacklists to detect malicious URLs can classify most of the known URLs but are poorly effective when processing newly generated ones. To forestall greater economic losses, it is imperative to exert a method that can classify URLs in a timely manner. To improve timeliness of detecting malicious URLs, we use machine learning algorithms to automatically classify URLs. In this article, we selected the experiment results of several common machine learning models on our data set as the baseline and compared them horizontally with the outcome of random forest classifier. After that, we optimize the classifier to make the random forest classifier to achieve the best outcome within the lowest complexity.