Alzheimer's disease, a type of dementia that prevails significantly in society, presents an unprecedented challenge for individuals and society. Its gradual progression is marked by memory loss and cognitive decline, necessitating early detection. While machine learning and several existing ABC-based feature selection models have been investigated for this goal, efficiency remains a major concern. This paper focuses on the pressing requirement for an effective and precise Alzheimer's disease prediction system that can reduce classification errors while improving model efficiency and minimizing complexity through feature selection techniques. It utilizes the Alzheimer's Disease Neuroimaging Initiative 2 (ADNI2) dataset incorporating 94 attributes with pre-processing steps, including label encoding and outlier detection via the Interquartile Range (IQR) method. The proposed model deploys a random forest classifier to enhance accuracy alongside iterative refinement techniques. The primary purpose is to improve accuracy in predicting diseases and minimize classification errors, consequently boosting efficacy. To accomplish these objectives, a Modified Artificial Bee Colony (MABC) algorithm has been suggested for feature selection while simultaneously addressing local optima problems and reducing complexity. The MABC algorithm's innovation lies in its utilization of multiple Gaussian Distribution-based local information sharing that provides feature subset evaluation with more robustness. Random forest classification was performed iteratively after every feature selection process to optimize results. Over 25 cycles with 49 ultimate selected features, the proposed model accomplished outstanding performance by yielding an accuracy of 98%, demonstrating a remarkable improvement from pre-existing methods. Such progressiveness pushes Alzheimer's disease prediction efficiency forward by lessening required resources through reduced positivity rates and fewer designated features towards remaining efficient at scale among healthcare providers in the remotest geographies. Consequently, this paper introduces implementing MABC algorithms for enhanced Alzheimer's disease prediction accuracy without compromising on system-wide cost-effectiveness values along with rapidity concerns that are critical factors affecting low-resource settings' healthcare provision decision-making processes, thus making it incredibly impactful work deserving significant attention.

ABC, Modified ABC, local information sharing, IQR, Random Forest, ADNII2.