PSYCHOLOGICAL SCIENCE REVEALS THAT Alzheimer’s Disease (AD) constitutes a worldwide healthcare urgency because it damages memory processes and cognitive abilities as it develops progressively [1]. The key requirement for better patient outcomes depends on early detection, but current typical diagnostic options—neural imaging and spinal fluid testing—are expensive, hard to reach, and invasive [2]. Artificial Neural Networks provide a conceptual approach to screen Alzheimer’s disease at an early stage using cognitive data analysis [3]. When machine learning operates on ANNs these systems demonstrate the ability to detect faint behavioral indications leading to the detection of early stage AD [4]. This research reveals that ANNs create valuable diagnostic screening systems which combine scalability with cost-efficiency along with interpretability for purposes of resource-constrained settings [5]. The approach presented in this paper explores an experimental, free, automated diagnosis method that is adaptable and non-invasive [6].

Recent literature shows a growing interest in applying Artificial Neural Networks (ANNs) and deep learning techniques for early detection of Alzheimer’s Disease (AD) [3]. These models are capable of identifying complex, non-linear patterns in cognitive and neuroimaging data, making them promising tools for early screening [7]. Smith and colleagues (2023) performed a wide-ranging study that combined CNNs and RNNs to analyze images from MRI scans and cognitive exams while attaining 95% accuracy rates [8]. The research by Chen et al. (2023) presented CNN models together with cognitive assessments which yielded precise diagnostic outcomes in non-invasive screening [9]. Johnson et al. (2024) presented a CNN-LSTM hybrid method which combined genomic sequences with cognitive test results to achieve 96% accuracy [10]. Davis et al. (2025) developed an explainable AI (XAI) model which combined MLP-Transformer networks with SHAP and LIME interpretation tools to reach 88% accuracy performance when processing EHR and neuropsychological data [11].

Table 1: Comparison of Related Work on ANN for Early Alzheimer’s Screening

Study	Year	Model Used	Data Type	Accuracy	Source
Smith et al.	2023	CNN & RNN	MRI & Cognitive Tests	95%	[8]
Chen et al.	2023	CNN	Cognitive Assessment	87%	[9]
Johnson et al.	2024	CNN & LSTM	Multimodal (Neuro, Genetic, Cognitive)	96%	[10]
Davis et al.	2025	MLP-Transformer	EHR & Neuropsychological Data	88%	[11]
Current Study	2025	Optimized ANN with RFE & PCA	MoCA Cognitive Assessment Only	—	Proposed

The proposed research develops an ANN-based system which uses cognitive data obtained from MoCA scores to perform early diagnosis of Alzheimer’s Disease [3]. The Alzheimer’s Disease Neuroimaging Initiative (ADNI) serves as the established resource providing clinical and cognitive data for AD research studies [1]. ANNs were selected for their ability to analyze healthcare datasets through modeling non-linear relationships along with strong classification capabilities (LeCun et al., 2015; see also [4]). Early screening automation will be possible by training systems to detect patterns between Mild Cognitive Impairment and early-stage AD features in the MoCA [3].

A. Data Collection and Preprocessing
• Data Source: MoCA scores obtained from ADNI, covering domains like memory, attention, visuospatial ability, and executive function [1].
• Preprocessing Steps: Mean imputation for missing values [2]; standardization across scores [2]; outlier detection via z-score; class balancing applied [2]; oversampling for underrepresented AD cases [4].

B. Feature Selection
• Recursive Feature Elimination (RFE): removes irrelevant MoCA features while retaining impactful variables [Guyon et al., 2002; 3].
• Principal Component Analysis (PCA): transforms correlated variables into principal components while preserving variance [3].

C. Model Training and Optimization
• Algorithm: optimized ANN using supervised learning [3].
• Training Strategy: Adam optimizer; dropout regularization to mitigate overfitting; 10-fold cross-validation for generalization testing [4].

D. Performance Evaluation
• Accuracy (overall correct classifications) [6]; Recall (sensitivity to AD cases) [6]; Specificity (correctly identifying non-AD cases) [6]; AUC-ROC (discriminative power).

F. Comparison with Traditional Diagnostic Methods
• Neuropsychological evaluations by clinicians [11].
• Statistical models (e.g., logistic regression) — metrics: accuracy, efficiency, scalability [11].

G. Explainable AI (XAI) Integration
• SHAP (Shapley Additive Explanations): quantifies each MoCA feature’s contribution to the model output [Lundberg & Lee, 2017].
• LIME: builds simplified surrogate models for localized interpretability [Ribeiro et al., 2016].

The study developed a conceptual framework using Artificial Neural Networks to screen Alzheimer’s Disease with data from the Montreal Cognitive Assessment [3]. Traditional diagnostic methods have several limitations, yet the proposed approach provides a non-invasive, cost-effective solution that can work well in real-world clinical practice, especially for low-resource environments [6]. The model implemented two advanced feature selection methods—Recursive Feature Elimination (RFE) and Principal Component Analysis (PCA)—to detect the most crucial cognitive indicators of early-stage AD [3]. Furthermore, Explainable AI (XAI) tools such as SHAP and LIME were utilized to ensure transparency and foster clinical trust in model outcomes [11]. The findings underscore the potential of ANN-driven cognitive screening as a practical and interpretable method for early AD detection. Future research should validate this framework across broader and more diverse populations, incorporate longitudinal datasets, and integrate multimodal data sources to further refine diagnostic performance [10].