Original title: Machine learning-based multiclass classification of cognitive decline using ACE-III assessment data: a comparative study with SHAP explainability
The researchers trained twelve computer methods to sort people into four cognitive groups using scores and background details from the ACE-III assessment. They used a technique called SMOTEA method that creates extra training examples for groups with fewer cases to reduce class imbalance. to help the methods learn from groups with fewer examples, then tested them across five rounds of held-out data. LightGBM had the highest average scores on several measures, but its lead was statistically clear over only one of the other methods. SHAPA method for estimating how much each feature contributes to a model’s prediction. showed which details most influenced predictions, but the results still need testing on independent groups.
Imagine sorting a pile of appointment cards into four boxes for people with different levels of memory concern. Some boxes have far fewer cards, and details like age or health history may help with a decision without explaining it completely. In this paper, the boxes are the four cognitive groups, the cards are assessment records, and the explanations show which details most influenced each sorting decision.
Clinicians need to distinguish normal cognition, subjective decline, mild impairment, and dementia—not just sort people into two groups. This model uses information from an existing assessment, so it could support early screening in primary care, but it still needs testing on independent patient groups before clinical use.
The study uses 140 participants and reports cross-validation results, not testing on an independent cohort. LightGBM’s macro-F1 advantage was statistically significant only against Logistic Regression, not the other ten classifiers. The abstract says external validation is needed before claiming clinical utility.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Machine learning-based multiclass classification of cognitive decline using ACE-III assessment data: a comparative study with SHAP explainability — Mohamed Sherif N. Abd Rabou, Heba M. Tawfik, Nabil A. Ismail et al. — Scientific Reports — doi:10.1038/s41598-026-71441-1
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper.