A landslide susceptibility map (LSM) is essential to determine the probability of landslide occurrence. The preparation of the susceptibility map is necessary to have reliable causative factors and an appropriately trained model. This study proposes a statistical feature selection of 18 landslide causative factors and a comparison of 15 machine learning algorithms’ performance to determine the optimal susceptibility model. To ensure that the factors are independent, the study uses statistical approaches, namely Pearson Correlation Coefficient and multicollinearity analysis. Based on the feature (causative factors) selection analysis results, the curvature factor was removed since it shows a high correlation with the plan and profile curvatures. Multicollinearity among all inferred factors has not been indicated. Moreover, the model fitting results show that the extra trees achieve the highest average accuracy score (ACC), with a value of 0.9859, followed by random forest (ACC = 0.9857), multi-layer perceptron (ACC = 0.9780), and decision tree (ACC = 0.9700).