Abstract: In recent years, the slope instability has brought immeasurable costs to production and life of human. As a result, it is essential to correctly understand, analyze, and design the slope reasonably, and implement appropriate protective measures to minimize the loss and harm caused by its instability. By far, slope stability can be investigated using theoretical analysis, numerical modeling and machine learning prediction, among them machine learning prediction has been the most encouraging one. Many studies have been performed using machine learning algorithms to predict the slope stability. However, these methods suffers from poor accuracy and poor generalisation capbility, so its real-life application has been limited. In the current study, a machine learning-based slope safety and stability evaluation system is established by integrating principal component analysis, parameter adjustment, and influence factor weight analysis. It is shown that PCA can reduce the dimensions of the input variables from six to three while retaining 80% of the information; however, at the cost of the model’s effectiveness. The random forest and XGBoost (eXtreme Gradient Boosting) learning algorithms can both be employed to develop effective evaluation models for slope safety and stability. The comparative analysis of algorithms’ prediction effects established XGBoost as the best evaluation model, which can achieve the average accuracy of 92%, precision of 91%, recall of 96%, and the area under the receiver operating characteristic curve (AUC) of 0.95. In addition, this study employs three types of test methods: the chi-square test, F test correlation, and mutual information method, meanwhile by calculating and visualizing the importance of influencing factors, the influence of unit weight, slope height, internal friction angle and cohesion on slope stability is demonstrated. It has been shown that the unit weight is the most influencing factor for the slope stability. Finally, the slope safety protection measures are proposed by combining the evaluation results with the actual project.