Comprehensive evaluation of machine learning models for predicting the cognitive status of Alzheimer's disease subjects and susceptible
Abstract Alzheimer’s disease is one of the greatest public health challenge of our time. This requires strong predictive models to tease apart what we can do to detect and introduce early-prevention. In the same vain, machine learning models could be effective on the prediction of cognitive status o...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Springer
2025-07-01
|
| Series: | Discover Data |
| Subjects: | |
| Online Access: | https://doi.org/10.1007/s44248-025-00068-w |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Alzheimer’s disease is one of the greatest public health challenge of our time. This requires strong predictive models to tease apart what we can do to detect and introduce early-prevention. In the same vain, machine learning models could be effective on the prediction of cognitive status of subjects from the real study, thus early-identifying possible deterioration of cognitive ability of individuals who are susceptible. The data we use is provided by the National Alzheimer’s Coordinating Center (NACC) and contain many clinical and demographic predictors by which we characterize and model the cognitive status of the subjects. Using a set of classical machine learning algorithms for predictive modelling (Random Forest, Gradient Boosting, XGBoost, Decision Tree, AdaBoost, Neural Networks, Extra Tree Classifier) and state of art methods such as sequential-attention based Tabent transfer learning, we explore the best performing models which are effective to predict the cognitive status of the subjects given certain clinical and other characteristics. We perform intensive hyper-parameter tuning for each model using grid search and randomized search to fine-tune the performance of these models. We train more than 900 different combination of feature selector, models and hyperparameters. At last, we assess the performance of the models on multiple performance measures (accuracy, sensitivity, specificity and area under the ROC curve (AUC)). These models were trained to classify the cognitive status classes (COGSTAT) and mild cognitive impairment (NACCMCII). These results show that the models can be not only highly accurate, but also par to the current best estimates published in the literature based on only the quantitative data (based on our knowledge), and can thus provide clinicians with an even more reliable tool to predict the cognitive decline. Overall, this work shows us how machine learning can aid in the understanding of Alzheimer disease and its identification by offering an important means to improve patient outcomes through early treatment strategies based on predictive analytics. |
|---|---|
| ISSN: | 2731-6955 |