With the rapid development of the Internet, the number of malicious URLs is growing, too. It brings great security threats to people. The current malicious URL detection methods based on machine learning don't select the features effectively, so it will directly affect the classification effect of the model. In order to avoid the problem of weak generalization ability caused by the training of small amount of samples, a large number of samples are used for experiments. In this paper, Random Forest is optimized based on feature contribution and hyperparameter optimization, and a large number of sample experiments show that the detection efficiency of the model has been significantly improved. The detection precision reaches 94.85%, and the AUC value reaches 96.51%. It has high application value for malicious URL detection.