<![CDATA[Purpose – This study aims to develop and evaluate a hybrid semantic-tabular machine learning framework for supporting the automated assessment of research grant proposals in the SRIKANDI Research Management System at Universitas Hasyim Asy’ari Tebuireng Jombang. Design/methods/approach – The study employed a computational experimental design using historical institutional proposal data from the SRIKANDI system. A total of 190 proposals were labeled based on the institutional LPPM scoring threshold, consisting of 107 approved and 83 not approved proposals. The proposed framework integrates semantic features extracted from proposal narratives using IndoBERT with structured tabular features, including document completeness, proposal score, text quality, budget information, research field, and proposer track-record indicators. The fused 786-dimensional feature representation was classified using a Random Forest model with balanced class weights. Model performance was evaluated using accuracy, precision, recall, F1-score, ROC-AUC, confusion matrix analysis, and SHAP-based explainability. Findings – The hybrid model achieved strong predictive performance on the holdout test set, with 96.55% accuracy, 96.54% weighted F1-score, 96.75% weighted precision, 96.55% weighted recall, and ROC-AUC of 1.0000. The model correctly classified all approved proposals and misclassified only one not-approved proposal. SHAP analysis showed that word count, reference count, proposal text quality, and proposal score were the most interpretable contributors, while IndoBERT semantic dimensions added meaningful predictive value beyond administrative features. Research implications/limitations – The findings indicate that hybrid semantic-tabular learning can support more consistent and transparent preliminary proposal screening. However, the study is limited by its single-institution dataset, relatively small sample size, high feature-to-sample ratio, and dependence on score-threshold labeling. Originality/value – This study contributes a replicable explainable AI framework that combines Indonesian-language semantic representation, institutional tabular features, and SHAP-based interpretability for research grant proposal assessment within a university research management system.]]>
