Background: Kidney clear cell carcinoma (KIRC) is the most common type of kidney cancer, accounting for approximately 60–85% of all the kidney cancers. However, there are few options available for early treatment. Therefore, it is extremely important to identify biomarkers and study therapeutic targets for KIRC. Methods: Since there are few studies on KIRC, we used a data-driven approach to identify differential genes. Here, we used miRNA gene expression profile data from the TCGA database species of KIRC and proposed a machine learning-based approach to quantify the importance score of each gene. Then, an ensemble method was utilized to find the optimal subset of genes used to predict KIRC by clustering. The most genetic subset was then used to classify and predict KIRC. Results: Differential genes were screened by several traditional differential analysis methods, and the selected gene subset showed a better performance. Independent testing sets from the GEO database were used to verify the effectiveness of the optimal subset of genes. Besides, cross-validation was made to verify the effectiveness of the approach. Conclusions: Finally, important genes, such as miR-140 and miR-210, were found to be involved in the biochemical processes of KIRC, which also proved the effectiveness of our approach.