Integrating Handwriting Analysis and Machine Learning for Enhanced Personality Trait Prediction

Abstract

   This thesis presents an in-depth exploration of graphology and its integration with machine learning to analyze personality traits through handwriting. The motivation for this research stems from the brain's ability to express personality traits through neuromuscular movements, particularly in handwriting. This study bridges the historical graphological methods, tracing back to the 19th century, with contemporary machine learning techniques.

This research utilized a dataset of 1,108 handwriting examples. CENPARMI contributed 234 of these, while the remaining 874 were procured through a business-oriented graphology expert. The data used comprises a diverse set of handwriting samples, analyzed using machine learning algorithms such as KNN, Random Forest, Logistic Regression, and specifically the VGG16 model for transfer learning. The research employs techniques like SMOTE for data balancing and ensemble methods for classification, including Majority Voting and Stacking Method.

Experimental results demonstrate a significant improvement in the accuracy of personality trait predictions after using SMOTE, with the highest accuracy exceeding 90% for traits like "Agreeableness" and "Open to Experience" using the Ensemble method (Stacking Method). Thus, this integrated approach allows the results with the enhanced predictive accuracy.

The main contributions of this research lie in its innovative integration of graphology and machine learning for personality assessment, methodological advancements in handling imbalanced datasets, and the application of transfer learning in handwriting analysis. The improved accuracy in personality trait prediction illustrates the potential of this interdisciplinary approach in fields such as psychology and personalized services, offering new insights into personality psychology and opening avenues for future research in this domain.