As science and technology continue to advance, more and more studies show that there are so many miRNA-disease associations (MDAs). Known MDAs will help us prevent and treat certain diseases. However, traditional wet experiments greatly consume manpower and time. Therefore, it is critical that a few reliable methods for predicting MDAs are developed. In this study, we have come up with a machine-learning method for quadratic prediction MDAs (MLQP). MLQP is an improvement method on the traditional least squares method. In this method, MDAs information can be used more by using Weight K Nearest Known Neighbors (WKNKN) method and neighborhood similarity processing (NSP). Gaussian interaction profile (GIP) kernel similarity is applied to improve the accuracy of prediction results. The regularization least squares method is also utilized to generate the prediction scores. AUC value of MLQP under five-fold cross-validation is 0.957. MLQP outperforms other methods in predicting MDAs, as demonstrated by the final experimental results. Finally, the efficacy and practicality of the MLQP will be further verified in the study of four specific diseases.