ABSTRACT
Handwriting is a complex neuromuscular activity that reflects an individual’s cognitive state, emotional tendencies, behavioral patterns, and personality traits. Traditional graphology relies on subjective human interpretation, which lacks consistency and scientific reproducibility. To overcome these limitations, this project proposes a hybrid AI-based personality prediction framework that integrates handcrafted graphology features with deep visual handwriting embeddings. Handwritten samples are processed through a Multi-Output Random Forest model for multi-trait prediction, while handwriting images are analyzed using a ResNet50-based CNN to extract stroke-level visual patterns. These numerical graphology features and CNN embeddings are fused and classified using an SVM to achieve highly accurate concentration prediction. A Tkinter GUI is developed to allow multi-image uploads, real-time processing, and clear visualization of all predicted traits. The system successfully transforms traditional subjective handwriting analysis into a data-driven, automated, and reliable method for personality assessment.
Keywords: Handwriting Analysis, Graphology, Personality Prediction, Random Forest, CNN, ResNet50, Hybrid Model, SVM, Machine Learning, Deep Learning.


OBJECTIVES
• To develop an automated AI-based handwriting personality prediction system that eliminates the subjectivity of traditional graphology.
• To extract and analyze core graphology features such as slant, margins, spacing, baseline alignment, and letter size from the handcrafted CSV dataset.
• To build a Multi-Output Random Forest model capable of predicting multiple personality traits simultaneously using structured handwritten features.
• To design a deep learning model using ResNet50 for extracting high-level visual handwriting embeddings from images.
• To combine CNN-based visual features with handcrafted graphology features through a hybrid feature fusion approach.
• To implement an SVM classifier over the fused features to achieve highly accurate concentration prediction.
• To ensure robust preprocessing, including cleaning, encoding, scaling, and augmentation, for improved model stability.
• To develop a Tkinter-based GUI that supports multi-image upload and displays all predicted traits in real time.
• To evaluate the performance of machine learning, deep learning, and hybrid models using accuracy, F1-score, and confusion matrices.
• To create a scalable and extendable personality analysis system that can accommodate additional traits or advanced AI models in future work.