Comparison Between Logistic Regression and Random Forest for Determining Factors of Intimate Partner Violence in Peru

Abstract

Intimate  partner  violence  is  a  social  problem  that  has  been  studied  by  different  researchers  to  determine  the  factors  that  influence  its  occurrence,  considering  different  environments, moments and locations. Sixty-eight point two percent (68.2%) of women have been victims of violence and 31.7% have been victims of physical aggression in Peru. The present research proposes  nine  models  based  on  logistic  regression  and  random  forest  with  variants  such  as   chi-square, Entropy and Gini, and three subscenarios out of five, ten and twenty variables that used the dataset of complaints registered in 2016 at the Ministry of Women. The best result of each subscenario was obtained, but finally the best model was that of twenty variables which used the random forest “feature selection” (Entropy) and the random forest model (Gini).