Enhancing Early Intervention and Patient Care through Machine

Learning in Alzheimer's Disease Prediction

C. Mallika

, J. Vanitha

, Visalatchy

, S. Selvaganapathy

and K. Kalavani

Department of Master of Computer Applications, E.G.S. Pillay Engineering College, Nagapattinam, Tamil Nadu, India

Sr.G.1 School of Computer Science and Engineering, VIT Vellore, Tamil Nadu, India

Keywords: Machine Learning, Alzheimer’s Disease, Predictive Modeling, Administrative Health Data, Risk Prediction.

Abstract: Alzheimer’s disease (AD) is a neurodegenerative disorder with a growing prevalence worldwide. The

application of machine learning (ML) for early AD prediction can enhance early diagnosis and patient care.

This study employs large-scale administrative health data from the Korean National Health Insurance Service

(NHIS) to develop predictive models for AD incidence. Three ML models Random Forest, Support Vector

Machine, and Logistic Regression were trained using 40,736 elderly individuals’ data with 4,894 unique

clinical attributes. The best-performing model achieved an area under the curve (AUC) of 0.775 for one-year

predictions. Key predictive features include hemoglobin levels, age, and urine protein levels. The study

highlights the potential of ML in AD prediction and its implications for early intervention.

1 INTRODUCTION

Alzheimer’s Disease (AD) is an age-related

neurodegenerative disease characterized with

progressive loss of cognitive function, including the

loss of memory, ability to reason, and perform daily

activities. The disease incidence is continuing to

increase and will be expected to worsen as the global

population is aging, leading to the emergence of AD

as a major health and healthcare burden for

caregivers. AD constitutes 60-70% of all dementia

cases globally and this highlights the imperative of

early diagnostic and treatment policies (WHO). (C

Mallika et al. 2024) The current diagnostic modalities

usually use neuroimaging tools like MRI/PET scans

and biomarker assessment from the cerebrospinal

fluid (CSF), that are effective but expensive and not

universally available.

One of the difficulties in detecting AD is the late

diagnosis – symptoms show several years after the

onset of the brain's pathological disease. Most of

patients are not diagnosed until the disease has spread

and invades into advanced stages, and treatment

strategies are of limited efficacy. (P Umamaheswari

et al. 2024) Hence, other methods are being

investigated to take advantage of massive health data

for recognizing pattern and AD risk for earlier

diagnosis. Machine learning is also promising with

regard to looking at complex datasets with the

potential for predictive models to evaluate risk of AD

using standard medical records, demographics, and

clinical history.

This study sets out to assess the ability of

machine learning models in predicting AD based on

administrative health data. 3 By integrating large

volume of de-identified patient records, it is possible

to evaluate historical health trends and to discover

major risk factors linked to AD. Our method can be

scaled in a cost-effective manner as a complement or

alternative to established diagnostic approaches that

could allow early detection and personalized

interventions at large scale. By carefully evaluating

models and comparing performance, this work can

inform the emerging field of AI-enabled healthcare

solutions, which aim to enhance patient outcomes and

reduce the societal impact of AD.

2 METHODOLOGIES

Data source We used the data from the NHIS

database (2002-2010), which is a deidentified patient

records of 40,736 elderly individuals (over 65 years

old). The dataset contains a variety of information

such as demography, medical history (including the

past history of disease, and laboratory test) and usage

Mallika, C., Vanitha, J., Visalatchy, , Selvaganapathy, S. and Kalavani, K.

Enhancing Early Intervention and Patient Care through Machine Learning in Alzheimer’s Disease Prediction.

DOI: 10.5220/0013913100004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 4, pages

361-364

ISBN: 978-989-758-777-1

361

history of drugs. The work refinements the early

prediction of AD through a diagnostic approach by

differentiating ‘definite AD’, with a formal diagnosis

and a dementia medication being prescribed from that

given for ‘probable AD’, where only a diagnosis but

not medication use is recorded.

Random Forest, Support Vector Machine, and

Logistic Regression had been used to develop

predictive models. (Wang et al. 2024). These models

were trained by sampling from both the balanced and

unbalanced version of the dataset to correct for

potential biases due to class distribution. Feature

selection method was performed to select the

important features of AD to make the models

interpretable and practical. Algorithm-specific

hyperparameters were tuned through training to

maximize predictive performance.

Various performance measures have been

considered for assessing the model such as accuracy,

precision, recall, along with the area under the AUC-

ROC curve for the receiver operating characteristic

(Mirabnahrazam, G.,et.al.,2022) (ROC). Cross-

validation methods were used to test the

generalizability and robustness of the model. The

discriminatory ability of each models predicting AD

incidence at different time points were examined,

demonstrating the promise of machine learning for

early disease detection and risk stratification. Figure

1 shows the pipeline for Prediction of Alzheimer’s

disease.

The Alzheimer’s Disease Neuroimaging Initiative

(ADNI) is a large data repository for multi-modal

neuroimaging scans, cerebrospinal fluid (CSF) and

genetic-based biomarkers, and clinical diagnosis,

suitable for training machine learning models to

predict AD. (Li, et.al.,2023) In this work, three

models, including RF, SVM, and LR, are trained on

ADNI data to screen for risk of AD. We then pre-

process the dataset, including normalizing the

measurements of neuroimaging, encoding the

categorical features, and normalizing the level of the

biomarkers, in order to normalize its format and

improve the performance of the model. (Mao

et.al.,2023) These features are fed into the models,

which are learned to separate AD from non-AD

based on the patterns observed in the data.

The ensemble learning model, Random Forest, is

well adapted for the high-dimensional medical data

like ADNI, and can combine several decision trees

together to decrease the variance and over-

fitting(Deepan et.al.,2023). In training, a set of

decision trees is constructed using random samples

from the data, and the predictions are made by using

majority voting. Hyperparameters (number of trees,

depth of tree and feature selection criterion) are tuned

to improve the accuracy. Likewise, SVM is utilized

to establish an optimal hyperplane which separates

the case of AD and non-AD. (Mallika et.al.,2022)

This model can use kernel functions like the RBF to

project the complex, non-linearly inseparable data in

to higher-dimensional spaces for better classification.

SVM hyperparameters are tuned using cross-

validation to achieve highly robust classification.

Figure 1: Pipeline for Alzheimer's Disease Prediction.

Logistic Regression (LR), as a probability model,

is used to explain the decision since it offers a better

interpretability, which provides clinicians with

insights into the contributions of each biomarker to

AD risk. (Mallika et.al.,2022) This model predicts the

risk of developing AD by multiplying input features

by weighted coefficients and applying a sigmoid

activation function to produce probability scores. We

conduct the feature selection to get rid of the

redundant variables and regularization methods to

avoid overfitting. (Park et.al.,2020) Model

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

362

performance is evaluated using metrics like

accuracy, precision, recall, F1-score, and AUC-ROC,

making good predictions. (Mallika et.al.,2019) The

optimal model is chosen for clinical deployment to

aid in early detection of AD and planning of

interventions.

Table 1: Performance Analysis of Machine Learning Models for AD Prediction.

Model Accuracy (%) Precision (%) Recall (%) F1-Score (%) AUC-ROC (%)

Random Forest (RF) 91.2 89.8 90.5 90.1 93.5

Support Vector

Machine (SVM)

88.6 86.3 87.9 87.1 90.8

Logistic Regression

(

)

85.4 82.7 84.1 83.4 88.2

3 INTERPRETATION OF

RESULTS

• Random Forest achieved the highest

accuracy (91.2%) and AUC-ROC (93.5%),

demonstrating strong overall performance.

(Mallika et.al.,2017)

• SVM performed well with an accuracy of

88.6% and a good balance between precision

and recall.

• Logistic Regression had the lowest

performance but remained interpretable,

making it useful for understanding feature

importance.

Figure 2: Performance Comparision.

Figure 2, here is a performance comparison of the

machine learning models (Random Forest, SVM, and

Logistic Regression) for Alzheimer's Disease

prediction. The chart visualizes the accuracy,

precision, recall, F1-score, and AUC-ROC.

4 CONCLUSIONS

In this paper we show the power of machine

learning models applied to ADNI dataset in terms of

AD prediction. Models (random forest, support

vector machine (SVM), logistic regression) were

again trained with different clinical and demographic

covariates with respect to AD risk. Among the

models, Random Forest model showed superior

accuracies, precision, recall, and AUC-ROC

compared with other models (Table 1), which

suggested that the model was highly robust on AD

prediction. The findings indicate that machine

learning may be useful as a screening tool for early

warning signs of PPCM, early identification of which

can allow prompt medical management and

treatment. But the performance of model is affected

by the quality of the data, feature selection and

imbalance of class, so the model needs to be

optimized.

5 FUTURE ENHANCEMENTS

More advanced deep learning models including

CNNs or RNNs can be studied in the future for better

prediction accuracy by associating intricate patterns

of neuroimaging and clinical data. Combining multi-

modal information such as genetic, brain images, and

lifestyle factors might improve predictive ability. We

argue that explainable AI (XAI) methods should also

be included to improve interpretability, and thus the

trust of clinicians in AI-based diagnoses. Developing

online predictive systems for clinical purposes may

be useful for guiding early interventions and

personalized treatment options. Moreover, increasing

sample size with more populations will enhance the

Enhancing Early Intervention and Patient Care through Machine Learning in Alzheimer’s Disease Prediction

363

generalization of the model and allow its greater

utilization in the clinical field.

REFERENCES

C Mallika, J Vanitha, K Kalaivani, S Selvamuthukumaran’

Big Data Analytics-Based Diabetes Prediction Model

for Identifying Internal Factors for Diabetes Mellitus”

2024 3rd Edition of IEEE Delhi Section Flagship

Conference

Deepan, S., et al. "The Role of Big Data Analytics in

Healthcare: Prospect and Ethical Consideration." 2023

10th IEEE Uttar Pradesh Section International

Conference on Electrical, Electronics and Computer

Engineering (UPCON). Vol. 10. IEEE, 2023.

Hashemifar, S., Iriondo, C., Casey, E., & Hejrati, M.

(2022). DeepAD: A Robust Deep Learning Model of

Alzheimer's Disease Progression for Real-World

Clinical Applications. ArXiv preprint arXiv:2203.090

96.

Li, R., Wang, X., Berlowitz, D., Silver, B., Hu, W., Keating,

H., Goodwin, R., Liu, W., Lin, H., & Yu, H. (2023).

Early prediction of Alzheimer's disease leveraging

symptom occurrences from longitudinal electronic

health records of US military veterans. arXiv preprint

arXiv:2307.12369.

Mallika, C., Selvamuthukumaran, S. Hadoop framework:

Analyzes workload predicition of data from cloud

computing ‘IEEE Xplore-2017

Mallika, C., and S. Selvamuthukumaran. "Privacy protected

medical data classification in precision medicine using

an ontology-based support vector machine in the

diabetes management system." Proc Int J Innovative

Technol Exploring Eng 9 (2019): 334À342.

Mallika, C., and S. Selvamuthukumaran. "Technological

perspective on precision medicine in the context of big

data a review." Proceedings of the International

Conference on Cognitive and Intelligent Computing:

ICCIC 2021, Volume 1. Singapore: Springer Nature

Singapore, 2022.

Mallika, C., Selvamuthukumaran, S. A Hybrid Crow

Search and Grey Wolf Optimization Technique for

Enhanced Medical Data Classification in Diabetes

Diagnosis System. Int J Comput Intell Syst 14, 157

(2021)

Mallika, C., and S. Selvamuthukumaran. "Hybrid Online

Model for Predicting Diabetes Mellitus." Intelligent

Automation & Soft Computing 31.3 (2022).

Mao, C., Xu, J., Rasmussen, L., Li, Y., Adekkanattu, P.,

Pacheco, J., Bonakdarpour, B., Vassar, R., Jiang, G.,

Wang, F., Pathak, J., & Luo, Y. (2022). AD-BERT:

Using pre-trained contextualized embeddings to predict

the progression from mild cognitive impairment to

Alzheimer's disease. arXiv preprint arXiv:2212.06042.

Mirabnahrazam, G., Ma, D., Lee, S., Popuri, K., Lee, H.,

Cao, J., Wang, L., Galvin, J.E., Beg, M.F., & the

Alzheimer's Disease Neuroimaging Initiative. (2022).

Machine Learning Based Multimodal Neuroimaging

Genomics Dementia Score for Predicting Future

Conversion to Alzheimer's Disease. arXiv preprint

arXiv:2203.05707

P Umamaheswari, C Mallika, M Vanitha, D Rubidha Devi,

P Dinesh, R ThanujaEarly Detection And Prediction Of

Sleep Apnoea Using Deep Learning Techniques” 2024

International Conference on Advances in Data

Engineering and Intelligent Computing Systems

(ADICS)-2024

Park, J.H., Cho, H.E., Kim, J.H., Wall, M.M., Stern, Y.,

Lim, H., Yoo, S., Kim, H.S., & Cha, J. (2020). Machine

learning prediction of incidence of Alzheimer's disease

using large-scale administrative health data. NPJ

Digital Medicine, 3(1), 46.

Wang, J., Ahn, S., Dalal, T., Zhang, X., Pan, W., Zhang, Q.,

Chen, B., Dodge, H.H., Wang, F., & Zhou, J. (2024).

Augmented Risk Prediction for the Onset of

Alzheimer's Disease from Electronic Health Records

with Large Language Models. arXiv preprint

arXiv:2405.16413.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

364