Implementation of the XGBoost Algorithm for Predicting Entrepreneurial Interest Categories of Accounting Department Students at SMK Islam Al Fadhila Demak

Abstract

Digital entrepreneurship offers vocational high school graduates a potential career pathway, yet schools have limited approaches for identifying individual students’ entrepreneurial interests. This study aimed to classify entrepreneurial interest as a proxy for digital entrepreneurial potential among accounting students at SMK Islam Al Fadhila Demak, identify influential predictors, and integrate the classifier into a prototype web system. A quantitative survey covered all 64 twelfth-grade students, assessing ten indicators through 30 Likert-scale items. Entrepreneurial interest scores of at least 13 defined the high-interest category (n = 20), while lower scores defined the low-interest category (n = 44). The remaining nine indicators served as predictive features. Extreme Gradient Boosting (XGBoost) was employed, with the synthetic minority oversampling technique (SMOTE) restricted to training data. Model development and evaluation involved stratified five-fold cross-validation and an 80:20 data split. The model achieved 84.49% accuracy, 84.79% macro precision, 80.56% macro recall, and an 80.92% macro F1-score. Accuracy exceeded the majority-class baseline of 68.75%, although recall for the high-interest category was 70.00%, indicating limitations in identifying students within this group. Risk-taking and digital marketing received the highest feature importance scores, at 52.25% and 17.15%, respectively. These importance values describe their contributions to model predictions and do not establish causal relationships. The results suggest that XGBoost can support exploratory classification of entrepreneurial interest within the studied cohort. However, the small sample, self-reported measurements, and use of shared folds for tuning and evaluation limit confidence in generalizability. The prototype has also not undergone user testing. Further validation with larger, independent samples and evaluation by intended users are needed before broader implementation. The proposed system offers a preliminary basis for student screening and prioritizing entrepreneurship coaching, while entrepreneurial interest remains distinct from demonstrated entrepreneurial capability.