<![CDATA[Timely graduation is a fundamental metric in higher-education accreditation and a key indicator of institutional efficiency. Conventional prediction models largely rely on longitudinal academic records, which are lagging indicators and often fail to detect risks during the early stages of study. This research proposes a paradigm shift by leveraging behavioral biometrics—specifically, the analysis of handwritten signatures using Deep Learning—to predict students’ graduation timelines and academic motivation profiles. Using a dataset from the Undergraduate Information Technology Program at Universitas Darma Persada, the study adopts the Cross-Industry Standard Process for Data Mining (CRISP-DM) framework. A Convolutional Neural Network (CNN) model based on the ResNet-50 architecture was developed, employing transfer learning to extract complex graphological features from signature images. Through rigorous data augmentation and statistical normalization, the model addresses the limitations of a small dataset. Empirical evaluation reports a graduation-prediction accuracy of 65% (Recall: 65%, F1-Score: 64%) and an academic-personality prediction accuracy of 70% (Precision: 74%, F1-Score: 69%). Although its absolute performance remains below transcript-based models, the findings validate the potential of signatures as early leading biometric indicators capable of capturing latent discipline and intrinsic motivation. This approach offers a non-invasive decision-support tool for academic advisors within intelligent education ecosystems.]]>
