Comparison of Classifiers Models for Prediction of Intimate Partner Violence

Vilma Susana Romero Romero, Victor Hugo Ayma Quirita, Juan Manuel Gutiérrez Cárdenas, Ashly Guerrero

Resultado de la investigación: Contribución a una revistaArtículo (Contribución a Revista)revisión exhaustiva

Resumen

Intimate partner violence (IPV) is a problem that has been studied by different researchers to determine the factors that influence its occurrence, as well as to predict it. In Peru, 68.2% of women have been victims of violence, of which 31.7% were victims of physical aggression, 64.2% of psychological aggression, and 6.6% of sexual aggression. Therefore, in order to predict psychological, physical and sexual intimate partner violence in Peru, the database of denouncements registered in 2016 of the “Ministerio de la Mujer y Poblaciones Vulnerables” was used. This database is comprised of 70510 complaints and 236 variables concerning the characteristics of the victim and the aggressor. First of all, we used Chi-squared feature selection technique to find the most influential variables. Next, we applied the SMOTE and random under sampling techniques to balance the dataset. Then, we processed the balanced dataset using cross validation with 10 folds on Multinomial Logistic Regression, Random Forest, Naive Bayes and Support Vector Machines classifiers to predict the type of partner violence and compare their results. The results indicate that the Multinomial Logistic Regression and Support Vector Machine classifiers performed better on different scenarios with different feature subsets, whereas the Naïve Bayes classifier showed inferior. Finally, we observed that the classifiers improve their performance as the number of features increased.
Idioma originalInglés estadounidense
Páginas (desde-hasta)469-488
Número de páginas20
PublicaciónAdvances in Intelligent Systems and Computing
DOI
EstadoPublicada - 1 nov 2020

COAR

  • Artículo de conferencia

Categoría OCDE

  • Ingeniería de sistemas y comunicaciones

Categorías Repositorio Ulima

  • Ingeniería de sistemas / Diseño y métodos

Temas Repositorio Ulima

  • Forecasting
  • Gender-based violence
  • Prospectiva
  • Violencia de género

Huella

Profundice en los temas de investigación de 'Comparison of Classifiers Models for Prediction of Intimate Partner Violence'. En conjunto forman una huella única.

Citar esto