A feasible method for training classified data with sparsity

Family harmony is an important part of social stability and harmony. Facing the rapid growth divorce disputes, how to hear cases quickly fairly urgent matter be solved. Based on judgment documents from courts located in Southwest China, this paper studies evaluate predict whether or not a case should divorced. These are characterized by their unbalance sparsity; besides that, most variables bearing with missing value. We propose feasible method that shows high accuracy both training testing datasets. Concisely, oversampling "clustering" exploited data pre-preparation, recursive feature elimination applied deal selection. In we combined Random Forest XGboost derive more precise model achieves ninety percent accuracy.