Early Alzheimer’s Prediction Using Dimensionality Reduction Techniques

Abstract

Single cell RNA sequencing (scRNA-seq) technology is capable of generating a large amount of data. These big data would be easier to handle, if they could be reliably reduced. In order to do so, many methods have been proposed. The dimensionality reduction (DR) and the feature Selection (FS) techniques proved to be favorable from the research field. Here the techniques mentioned above are reevaluated and seven different Machine Learning (ML) methods are identified aiming to present a comparison of the selected methods according to the model performance, metrics, and speed. Alzheimer’s disease damages the brain and is the most frequent neurodegenerative illness among older people. Alzheimer’s disease causes a gradual loss of cognition and memory. Genetic factors play a significant influence in the start of the disease, as certain genes might exacerbate suffering while not directly causing the disease. In recent research, dimensionality reduction and feature selection have been used in Alzheimer’s disease in order to improve the classification performance. When the number of observations is much lower than the dimensionality of the features, unsatisfactory classification performance results. The “small sample size” problem refers to the problem of poor classifier performance. Using manifold learning approaches, the dimensionality of the feature vector can be decreased. As a result, managing shape data in a low-dimensional space becomes more tractable. The most useful information could be kept utilizing dimensionality techniques, offering enhanced performance. The applied information gain approach to optimize feature selection in machine learning reveals that with optimized feature selection, classification accuracy improves, indicating that the information gain method can be used to choose more sensitive anatomical regions in Alzheimer’s disease diagnosis.