Machine Learning Algorithms for ccRCC Data Analysis

Malignant tumors are one of the top ten causes of death worldwide. Clear cell renal cell carcinoma (ccRCC) is the most common renal cell carcinoma. Discovering latent factors through genetic testing help diagnose the disease. Due to the complexity of proteomics, we used machine learning models to classify the disease and choose proteins with important features, in order to reduce irrelevant detection. We used different machine learning methods, including logistic regression (LR), support vector classification (SVC), and random forest classifier (RFC) for feature selection with different classification algorithms, e.g., SVC, XGBoost, and Multilayer Perceptron, to predict diseases. Differences among various machine learning methods were compared to improve classification accuracy and present key features. The ccRCC dataset with 11,817 proteins and 194 patients was used for model development. According to the evaluation, LR and RFC showed a larger area in the receiver operating characteristic curve (AUC), compared to SVC. In the LR ccRCC dataset, the feature selection method achieved an AUC of 0.995 while the RFC feature selection method of RFC achieved AUC up to 0.996. By comparing different classification algorithms with the RFC feature selection method, RFC, XGBoost, and LR achieved AUCs of 0.996, 0.995, and 0.992, respectively, where LR has the lowest AUC score. However, using the LR classifier model with the LR feature selection method achieved an AUC of 0.995. These results demonstrated that different machine learning algorithms must be matched with different feature selection methods to classify data.