Xiaohong Yin ¹ Mengru Ba ¹ Qingguo Ren ² Shaoyuan LI ^1*

• ¹ Qingdao University of Science and Technology, Qingdao, China
• ² Qilu Hospital of Shandong University (Qingdao), Qingdao, Shandong, China

Parkinson's disease (PD) is a neurodegenerative disorder prevalent among the elderly, often leading to cognitive impairment in severe cases. However, its early stages may lack significant non-motor symptoms, resulting in a high misdiagnosis rate. To promote early treatment and targeted intervention for mild cognitive impairment (MCI) in PD patients, this study proposes a Stacking ensemble model with Optuna hyperparameter optimization. The model features a two-layer learning framework, comprising a basic classifier learning layer and a metamodel learning layer, aimed at improving the accuracy of predicting MCI in PD patients. Firstly, we designed a feature selection method based on multimodal data, identifying seven risk factors closely associated with cognitive impairment. Secondly, we constructed the Stacking ensemble model using five different learning methods to build the base classifier learning layer, with Extreme Gradient Boosting (XGBoost) serving as the metamodel. Additionally, the key parameters of XGBoost were autonomously optimized using the Optuna hyperparameter optimization method to enhance the predictive accuracy of the proposed Stacking ensemble model. Experimental results demonstrate that the proposed Optuna hyperparameter-optimized Stacking ensemble method exhibits high accuracy and effectiveness, with an accuracy of 0.9200, precision of 0.8768, recall of 0.9495, and an F1 score of 0.91315.