Alzheimer's disease (AD) is the leading cause of dementia worldwide, indeed, the world health organization estimates the number of dementia cases to reach 153 million by 2050 world widely. Conventional diagnostic pathways, such as neuroimaging, cerebrospinal fluid biomarkers, and specialist neuropsychological batteries require infrastructure that is largely unavailable in low and middle income countries (LMICs), where the median interval from symptom onset to diagnosis exceeds 3 years. This paper presents a preregistered protocol and literature-derived findings for a low-cost, dual inertial-measurement-unit (IMU) gait-screening framework for early-stage AD in community-dwelling adults aged 65 years and above. Two MPU6050 (HW-290) accelerometers, placed at the lateral malleolus and the lumbar L5 vertebra, were used to record 20-metre instrumented walks from 200 age-stratified participants (100 early-stage AD and 100 healthy controls). Nine spatiotemporal gait features including walking speed, cadence, step length, step time, foot clearance, frontal and lateral sway, and step/stride length variability were extracted using a validated peak-detection pipeline, and three supervised classifiers (Support Vector Machine (SVM), Random Forest (RF), and a Multi-Layer Perceptron (MLP)) were benchmarked under nested cross-validation. Effect sizes reported in the literature ranged from moderate (Cohen's d = 0.59) to large (d = 0.98), with step length variability and walking speed as the strongest discriminators, and a classification AUC-ROC of 0.89 was reported in the literature. The proposed protocol establishes a reproducible, ethically governed methodology for a screening instrument deployable at the community-health-worker level without specialist infrastructure, resulting in improving the public health and life expectancy.

Alzheimer's disease, gait analysis, inertial measurement unit, low-cost screening, machine learning classification, underserved communities.