Development of a stacking-based ensemble machine learning for detection of depression in Parkinson's disease

The primary symptom of PD is dyskinesia, but as Parkinson's disease (PD) progresses, non-motor symptoms such as depression are more likely to occur. According to previous studies, even though patients with PD frequently experience depression, only 1% of them recognized depression by themselves. These results suggested that it would be needed to diagnose Parkinson's disease depression as soon as possible. Previous studies that evaluated the factors related to Parkinson's disease depression in South Korea reported that neuropsychological level, health factors, socioeconomic status, education level, age, spouse, and social activities affected Parkinson's disease depression. Since regression analysis was mainly used as a modeling method to predict depression, they were efficient in identifying individual risk factors. However, they were limited in identifying compound-risk factors such as sociodemographic variables and living habits. Moreover, since regression analysis assumes independence, normality, and homoscedasticity, there is a possibility of producing biased results when the model is developed using data in violation of normality. As a way to overcome the limitation of the regression model, Classification is a machine learning technique has been widely used in Clinical decision supporting system and medical artificial intelligence. Machine learning can analyze data accurately even if the data somewhat violate the assumption of normality such as nonlinear data in the estimation process. In particular, Classifier Ensemble has a better accuracy than a single classifier, so active research has recently been conducted. The objectives of this study were to develop a model for predicting Parkinson’s disease depression based on stacking-based ensemble machine learning. This study was conducted with resources of'Parkinson's Disease Epidemiology Data' from National Biobank of Korea, the Centers for Disease Control and Prevention, Republic of Korea. This study analyzed 280 subjects who were 60 years or older with Parkinson's disease. Depression was measured using 30 items of Geriatric Depression Scale (GDS). This study was combined with base learners by stacking, in which a meta-learner (meta-classifier) was served to combine the predictions of base learners. A random forest was determined to play the role of meta-learner. This study compared the prediction performance of each model and determined that a model with the highest accuracy with 0.6 or higher sensitivity and specificity as the best model. If models have the same accuracy, the model with the high sensitivity value was selected as the best prediction model. This study suggests that stacking-based ensemble machine learning may be more effective in predicting Parkinson's disease depression than a single classifier.