Page 128 - DP-2-3

Design+ ML for predicting Alzheimer’s progression

critical role of cognitive assessments in the early detection Ultimately, this framework serves as a foundation for
and diagnosis of AD. developing intelligent, personalized diagnostic support
systems that prioritize early intervention and optimized
7. Conclusion and future work resource allocation.

In this study, non-imaging data from the AIBL were Given the time constraints of this research, explicit
analyzed to classify AD into three classes: HC, individuals handling of outliers, multicollinearity, or distribution
with MCI, and those diagnosed with AD. Extensive abnormalities was not performed. However, the selected
data cleaning and exploration were conducted to reveal models possess built-in capabilities to address these
underlying patterns and extract information using various issues. Involving domain experts to directly address
data mining techniques and statistical methods, including these factors could further enhance model performance.
correlation analysis, feature association analysis, feature Moreover, the models were fine-tuned using the
importance analysis, and feature selection. “RandomizedSearchCV” function; however, a more
28
To address the challenge of class imbalance, synthetic exhaustive approach, such as “GridSearchCV,” could
oversampling methods were employed to generate artificial potentially yield better parameters by exploring a wider
samples to balance the target classes. Subsequently, the data range of combinations. Although this study focused solely
were modeled and evaluated using advanced non-parametric on direct multi-class classification, alternative approaches
29
ML algorithms, such as RF and XGBoost, first with complete such as one-versus-one and one-versus-the-rest may
feature set and then with selected features obtained through offer additional insights. Acknowledging these limitations
the feature selection process. Fine-tuning techniques were provides a pathway for future research and further
applied to enhance predictive accuracy. The results from exploration of the problem.
these models underwent thorough evaluation to determine
the most effective algorithm. The tuned RF model emerged Acknowledgments
as the top performer, achieving an accuracy of 90%, with None.
precision, recall, and F1 scores also reaching 90%.
Furthermore, to reduce the diagnosis cost of AD and Funding
provide valuable insights toward developing a more None.
reliable and affordable diagnostic tool, the data were
segmented into three main variable groups: medical Conflict of interest
history, neuropsychological assessment, and blood analysis The authors declare that they have no competing interests.
with ApoE genotype variables. Corresponding ML models
were then developed using the fine-tuned model of the Author contributions
best-performing algorithm. The “neuropsychological Conceptualization: All authors
assessment” classifier emerged as the most effective, Formal analysis: Adhinrag Kalarikkal Induchudan
exhibiting an exceptional accuracy of 90%.
Investigation: All authors
Beyond its strong classification performance, this study Methodology: All authors
presents a replicable and cost-effective methodology that Writing–original draft: Adhinrag Kalarikkal Induchudan
may benefit other research groups, clinical practitioners, Writing–review & editing: All authors
and public health systems aiming to improve early detection
of AD. By leveraging non-imaging data—including widely Ethics approval and consent to participate
available neuropsychological assessments—our approach Not applicable.
avoids the high costs and limited accessibility associated
with imaging-based diagnostics. The use of interpretable Consent for publication
ML models, combined with robust feature selection and
data preprocessing techniques, facilitates deployment in Not applicable.
diverse clinical or research settings without the need for Availability of data
extensive computational infrastructure. Moreover, the
proposed methodology can be adapted to other populations Data is available from the corresponding author upon
or datasets, supporting generalizability studies and cross- reasonable request.
cohort validation efforts. This makes it particularly relevant
for low-resource settings or large-scale screening efforts References
where rapid, accurate, and affordable tools are essential. 1. Alzheimer’s Association. 2023 Alzheimer’s disease facts and

Volume 2 Issue 3 (2025) 10 doi: 10.36922/DP025270031