Pemetaan Kepribadian RIASEC melalui Klasifikasi Multi-Task Fitur Grafologi Tulisan Tangan Menggunakan ResNeXt50

Abstract

Personality analysis based on graphology—the study of handwriting characteristics to identify individual personality traits—is an approach that has been increasingly developed in the fields of psychometrics and artificial intelligence. This research proposes a method for mapping Holland’s personality types (RIASEC) through graphology-based handwriting analysis using a deep learning approach. Conventional personality assessments generally rely on self-assessment questionnaires, which are highly subjective. To address this limitation, this study develops a Convolutional Neural Network (CNN) model with a ResNeXt50 architecture based on multi-task learning to classify five graphological features: letter size, writing slant, word spacing, line spacing, and pen pressure. The dataset used in this study was obtained from the IAM Handwriting Database, consisting of 1,533 handwriting images. The data underwent preprocessing steps—including resizing, conversion to tensor format, and normalization—before being trained using a multi-head CNN model with cross-entropy loss for each graphological feature and the Adam optimizer for optimization. After the training process, the model was evaluated using a testing set that had never been used during the training or validation stages to objectively assess its generalization capability. The evaluation results indicate that the proposed model can classify graphological features with an average accuracy above 80% and map the classification results to RIASEC personality types with up to three dominant types. These findings indicate that the ResNeXt50-based multi-task learning approach has the potential to serve as a more objective, efficient, and applicable alternative method for personality assessment in the contexts of career development and education.