Handwriting-based personality classification on Indian samples using long-short term memory

Abstract

Traditional handwriting analysis methods have historically faced criticism for their lack of scientific basis, but more contemporary models based on layered artificial neural network (ANN) architecture have evidently been more successful. In the proposed model, a deep neural network (DNN) layered, long-short term memory (LSTM) model with contextual analysis has been proposed for handwriting-based personality classification. The model has been trained over a manually curated verbose dataset of ~6,000 Indian handwriting sample images, varying across genders, age groups, and regions. The classification is based on the five major personality traits. The proposed framework achieved an accuracy of 97.75%, which is over 10% better than the next best performing model on a comparably numerically bigger dataset; demonstrating the enhanced potential of context based neural networks on handwriting-based personality prediction when coupled with an appropriately varied and unbiased dataset.