Handwriting-based gender classification using machine learning techniques

Biometrics encloses the science of measuring human body characteristics and authorizing the user based on biometric modalities. Physiological and behavioural biometrics identifiers are the two kinds of biometrics. The present article i.e. Gender Classification system (GCS) is one of the most challenging and serviceable application among the many behavioural biometric based artificial intelligence and machine learning systems. In the current experiment, handwriting modality has been in practice for the gender classification. For the experimental evaluation, a corpus consisting of 200 writers with offline handwriting samples from 100 males and 100 females in Gurumukhi script has been pre-processed using pre-processing algorithms, followed by extracting features by exploiting Zoning, Diagonal, Peak extent, Transition and hybridization of feature extraction algorithms. Based on the extracted features, gender is classified using classification techniques, namely, K-NN, Decision trees, Random Forest, and Adaptive boosting methodology. Performance of the experiment has been analysed using evaluation metrics such as classification accuracy, precision rate, area under the curve, root mean square error and false-positive rate. The proposed system achieves maximum accuracy of 94.6% for gender classification using hybridization of features based on offline handwriting in Gurumukhi script.