Machine learning models operating on gene family data can require significant computational resources due to the large number of features, resulting in long training times and much computational time. Feature selection not only helps to reduce computational costs and speed up the processing but also minimizes the removal of important features that affect the prediction results. In this study, we evaluated the effect of the Local Interpretable Model-agnostic Explanations (LIME) integrated with Random Forest (RF) on gene family selection to perform disease prediction. The experimental results on the gene family datasets show that the classification algorithm can perform better on datasets with only 100 features than on original datasets with more than a million features. In addition, compared to previous methods, such as principal component analysis and random projection, we also obtained better classification performance, showing that integrating LIME and RF can be an effective gene family selection to improve disease prediction performance.