Novel use of logistic regression and likelihood ratios for the estimation of gender of the writer from a database of handwriting features
Vishal Sharma, Manjot Bains, Rajesh Verma, Neha Verma, Raj Kumar
Abstract
The present study focuses on the identification of individual characteristics in handwriting for the purpose of estimating the gender of the writer. A total of 150 individuals participated in the study and the relevant features were extracted from their handwriting. These characteristics were collectively studied and their potential of gender estimation was explored by utilizing statistical methods of chi-square test, logistic regression and likelihood ratios. First, the chi-square test was performed and it was observed that out of 25 individual characteristics, only 15 showed significant differences between males and females (p < .05) handwriting. Further, the logistic regression model reduced the number of such characteristics to 9, to cost-effectively assess the gender of the writer, in the process including 4 features that had p > .05. The correct classification rate for females and males thus arrived at was 80% and 76.4% respectively. Further, the likelihood ratio approach was also applied to assess the strength of evidence in the estimation of sex of the unknown person from these handwriting characteristics. Knowing the gender of the writer can establish priors for the statistical approach to the identification of the writer.