Alzheimer's disease (AD) is the most common type of dementia, a well-known term for memory loss and other cognitive disabilities. The disease is dangerous enough to interfere with daily life. Identifying AD in the early stages is a very challenging task, meanwhile the progression of it starts several years before noticing any symptoms. The main issue faced during diagnosis is high dimensionality of data . However, not all features are relevant for solving the problem, and sometimes, including irrelevant features may deteriorate the learning performance. Therefore, it is essential to do feature reduction by selecting the most relevant features.
In this work, a hybrid approach for high dimension feature selection is proposed. The dataset created by Alzheimer's Disease Neuroimaging Initiative (ADNI) was used for this purpose. The ADNI dataset contains 900 patients whose diagnostic follow-up is available for at least three years after the baseline assessment. This approach combines two well-known approaches, random forest and partial swarm optimization. Those approaches were chosen for their strength in solving large scale optimization problems with high data dimentionality. The Experiments show that our approach outperforms most of other approaches found in literature. It achieved high performance compared to them. The accuracy rate of this approach reached 95% for all the AD stages.