Augmented Backpropagation Neural Networks for Handwriting Analysis Based on Multi-Task Learning for Detecting Personality Attributes

Abstract

Handwriting analysis has been explored as a potential indicator of personality traits, though traditional approaches remain subjective. Many prior researches have shown a correlation between personality and handwriting variances, although the data is mostly anecdotal. Each individual’s handwriting is distinctive which may serve as indicators of underlying personality traits, behavioural patterns and psychological characteristics. This study proposes a machine learning-based framework using a Modified Backpropagation Neural Network (MBPNN) integrated with multi-task learning to objectively identify personality traits from handwriting samples. Feature extraction is performed using Discrete Wavelet Transform (DWT), enabling effective representation of handwriting characteristics. The incorporation of multi-task learning enhances the network's ability to generalize across related prediction tasks while minimizing computational overhead compared with training separate single-task models. The findings suggest that integrating DWT-based feature extraction with an optimized neural learning architecture provides an accurate, robust, and computationally efficient solution for automated handwriting-based personality assessment. The proposed model achieves an accuracy of 91% in multi-trait personality classification, demonstrating improved performance over conventional approaches. The results highlight the capability of the proposed method to extract robust features and provide reliable personality predictions.