Explainable Machine Learning for Analyzing AI Adoption and Teachers’ Professional Development Needs: Evidence from TALIS 2024 in Kazakhstan

The growing use of artificial intelligence (AI) in education highlights the need to distinguish teachers’ AI adoption from their professional development needs. This study examined these outcomes as separate predictive tasks using OECD TALIS 2024 data for 7360 teachers in Kazakhstan at ISCED level 2. After excluding invalid outcome records, the analytical samples comprised 2430 teachers for AI adoption and 4868 for professional development needs. Population estimates and confidence intervals were calculated using teacher weights and 100 Fay-BRR replicate weights, while predictive performance was assessed using out-of-fold predictions and school-grouped validation. The weighted prevalence of AI adoption was 59.0% (95% CI: 56.1–61.9). Logistic Regression and CatBoost demonstrated comparable discrimination (ROC-AUC = 0.754 and 0.753, respectively), with no clear performance advantage for CatBoost. SHAP analysis identified prior AI-related professional learning, uncertainty regarding AI attitudes, and perceived AI benefits as prominent predictive factors. Moderate or high professional development need was estimated at 80.3% (95% CI: 78.7–81.9). However, individual prediction was weak, with balanced accuracy of 0.268 and a quadratic weighted kappa of 0.082. These findings indicate that TALIS variables provide a moderate predictive signal for AI adoption but are insufficient for reliable, individualized professional development decisions.