Prediction Model of Postgraduate Innovation Capability Evaluation Results Using Random Forest Based on Hyperparameter Optimization Algorithms

The evaluation of postgraduate innovation capabilities holds significant importance in higher education. This study proposes a prediction model for assessing graduate innovation capabilities using the Random Forest (RF) algorithm. To balance model accuracy and computational efficiency, grid search (GS) and Bayesian optimization (BO) algorithms are employed for hyperparameter optimization. A comparative analysis with the Backpropagation Neural Network (BPNN) model is conducted to identify the optimal model. Results indicate that the RF model achieves the highest <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}$</tex> value using the Intellectual Innovation <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(U_{1})$</tex> feature set, while RF models under the Accumulation of Basic Knowledge for Innovation <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(U_{3})$</tex> feature set show the lowest MAE, MAPE, and RMSE values. Leveraging BO's advantage in low-dimensional parameter tuning, the optimized BO-RF model achieves the highest <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}$</tex> value (0.695) and lowest prediction errors (MAE <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$=7.712$</tex>, MAPE = 14.42%, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{RMSE}=9.731$</tex>), with a hyperparameter optimization time approximately 1/121 of GS-BPNN. This study provides a scientifically effective method and foundation for accurately predicting the future development of graduate innovation capabilities.