In recent years, miRNAs have been shown to play an important role in many diseases, most notably cancers. Traditional experiments for cancer diagnostics are time-consuming and expensive, thus more attention is being diverted towards discovering computational methods for predicting miRNA–disease association and using them in cancer diagnostics. miRNA numerical sequence information and genes targeted by miRNA were used for the construction of descriptors for machine-learning models. Next, we generated a table of miRNA descriptors using all of the miRNAs in a specific cancer dataset for disease classification. To show the effectiveness of the system, we constructed miRNA descriptor systems for pancreatic cancer, lung cancer, and breast cancer. With the Random Forest classifier, we obtained classification accuracies of 86.9%, 86.3%, and 85.1% for the above-mentioned cancers, respectively. Next, different disease datasets were tested on each model, including new miRNA sets for each cancer type from other studies. The models were able to classify the corresponding cancer miRNAs with >90% accuracy and other disease and cancer datasets with <60% accuracy. With this information, we constructed a hard-voting scheme using the three cancer classification models that is able to perform cancer diagnostics. The results suggest that our method is effective in miRNA–disease association prediction and performing cancer diagnostics. Full article