Software Engineering Development of Machine Learning Approaches for Alzheimer’s Disease Classification

Alatrany, A ORCID: 0000-0002-4504-1506 (2025) Software Engineering Development of Machine Learning Approaches for Alzheimer’s Disease Classification. Doctoral thesis, Liverpool John Moores University.

Preview

Text 2025AlatranyPhD.pdf - Published Version Available under License Creative Commons Attribution Non-commercial. Download (3MB) | Preview

Abstract

Alzheimer’s disease (AD) is a progressive and degenerative neurological disorder that profoundly impacts daily life. As the most common form of dementia, accounting for up to 80% of all cases, AD is marked by a gradual decline in memory, thinking, and behav- ior. What often begins with mild symptoms progresses to severe cognitive and physical impairments that compromise independence and quality of life. Despite affecting more than 55 million people worldwide, the precise causes of AD remain unclear and no cure currently exists, though treatments can help manage symptoms and slow progression. This thesis investigates the classification and prediction of AD by applying machine learning (ML) and data analytics techniques to genetic and multi-source datasets. A key challenge in AD research lies in the immense size of genetic data, which makes anal- ysis computationally intensive. To overcome this, transfer learning is introduced—an approach not previously applied in this domain. Convolutional Neural Networks (CNNs) were first trained on genome-wide association study (GWAS) data from the Alzheimer’s Disease Neuroimaging Initiative, and deep transfer learning was subsequently used to refine the model with a separate AD GWAS dataset. The final feature set extracted from this process was classified using a Support Vector Machine, achieving an accuracy of 89% and demonstrating the effectiveness of the proposed strategy.
Beyond predictive accuracy, high-dimensional data raises challenges for interpretabil- ity. To address this, the thesis develops a hybrid feature selection method combining association testing, principal component analysis, and the Boruta algorithm to iden- tify key predictors of AD. The selected features were then applied to wide and deep neural network models, which maintained high accuracy despite the dimensionality reduction highlighting the robustness of the approach.
Expanding beyond genetic data, a further methodology was applied using the multi- source dataset from the National Alzheimer’s Coordinating Center was analysed, en- compassing 45,923 participants, 1,023 variables, and 169,408 records across baseline and follow-up visits. Using the Boruta algorithm, a relevant subset of features was extracted, and among the tested classifiers, Random Forest achieved strong and balanced perfor- mance.
Finally, recognising that the “black-box” nature of ML models can limit clinical adop- tion, this work emphasises interpretability. Extended experiments uncovered meaningful patterns and risk factors for AD, with the Clinical Dementia Rating tool emerging as a particularly significant predictor. These findings not only strengthen the predictive framework but also provide clinically relevant insights into AD progression and risk profiling.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	Alzheimer’s disease; Machine learning; Genetic data; GWAS
Subjects:	B Philosophy. Psychology. Religion > BF Psychology Q Science > QA Mathematics > QA75 Electronic computers. Computer science R Medicine > RA Public aspects of medicine > RA0421 Public health. Hygiene. Preventive Medicine
Divisions:	Computer Science and Mathematics
Date of acceptance:	12 September 2025
Date of first compliant Open Access:	16 October 2025
Date Deposited:	16 Oct 2025 12:24
Last Modified:	16 Oct 2025 12:24
DOI or ID number:	10.24377/LJMU.t.00027200
Supervisors:	Al-Jumeily, D and Hussain, A
URI:	https://researchonline.ljmu.ac.uk/id/eprint/27200

View Item