Evaluation Of Feature Selection for Improvement Backpropagation Neural Network in Divorce Predictions

Abstract

In every domestic life always has its own problems, and every household must have their own conflict. Problems and conflicts that come in the life of the household can actually be part of the process to mature each other's partners, but sometimes the problems and conflicts that trigger divorce. Dataset from the UCI Repository, an evaluation and improvement process was carried out on the Backpropagation Neural Netwok(BPNN) algorithm and also performed a set of parameters are set and then validated, able to predict whether a married couple will divorce or not with sufficient results. high. And also do a process for some feature selection, then the value or rating of the most significant features will be processed on the Backpropagation Neural Network Algorithm and compare the results of the Gain Ratio, Information Gain, Relief and Correlation feature selection. This model underwent several validation processes so as to achieve a fairly high accuracy by using the Relief feature selection that is 99.41%.