Hybrid Features for Gender Classification System based on Pre-segmented Handwritten Gurumukhi Characters Using Machine Learning Algorithms

Abstract

Identification of handwriting has been an evolving task for mitigating specific issues in the fields of machine learning and pattern recognition. In the fields of artificial intelligence and machine learning, a gender classification method based on one or more features has emerged as a promising idea for addressing issues such as criminal, forensic, and suspected examinations. Predicting an individual's age, gender, and handedness based on their handwriting is one of the more challenging offline handwriting applications. The authors have presented a novel approach to classifying gender as male or female based on pre-segmented Gurumukhi (Punjabi) characters using feature extraction methods and classification algorithms such as K-NN, random forest, SVM, and MLPs in this proposed work. On a dataset of about 72,000 Gurumukhi characters, we extracted diagonal, transition, zoning, and peak extent based features (which consists of 280 writers with 150 males and 130 females). Our results show that by using a hybridization of multiple feature extraction techniques, the random forest classifier achieves higher accuracy than other classifiers.