Writer Trait Identification from Hindi Handwriting: A Hybrid Framework Combining Traditional And Deep Learning Models

Handwriting offers a unique behavioral biometric that can reveal critical information about a writer's identity and psychological state. This ‎study presents a hybrid classification framework for writer identification using Hindi handwritten text, with a focus on predicting age group, ‎gender, and anxiety level. A custom dataset was constructed containing diverse handwriting samples enriched with demographic and emotional metadata. The proposed system integrates both handcrafted features (HOG, ORB, LBP, SURF) and deep features extracted using ‎EfficientNet, evaluated using standalone classifiers (KNN, SVM), a hybrid ensemble (SVM + KNN), and a custom Convolutional Neural ‎Network (CNN). The hybrid models showed significant improvement over traditional classifiers, demonstrating the effectiveness of combining multiple feature representations. The CNN model outperformed all others, achieving accuracies of 83.7% for age prediction, 86.7% ‎for gender classification, and 76.8% for anxiety estimation. These findings validate the proposed approach as a robust solution for handwriting-based personal trait identification and open new avenues for intelligent, language-specific biometric systems in real-world applications ‎such as forensics, education, and psychological assessment‎.