<p>Feature selection is an area of machine learning techniques that has not been explored much. It reduces the number of insignificant features and the size of training sets. Many studies focus on features with numbers. However, features can also be categorized (such as by color or type), and these categories are often related to other features. When we turn categories into numbers, we lose these relationships. Therefore, we have proposed a feature ranking method based on feature probability estimation. This work presents a method for selecting significant features from a training set. It reduces computational and storage complexities by eliminating insignificant features. This method is applicable to both numerical and categorical data. The FPE enhances the system’s reliability and execution speed by removing unnecessary features. We have executed the proposed method on seven different datasets and compared it to popular feature selection techniques, including PCA, K-best (Chi-Square), Feature Importance, Information Gain, Mutual Information, Correlation, and Recursive Feature Elimination. The experimental results demonstrate that the proposed method outperforms existing feature selection methods in terms of speed and achieves higher accuracy than many of them; its performance is also comparable to that of the best feature selection method.</p>