A Comparative Study on Fisherface Techniques DLDA, PDLDA to Avoid SSS Problem

Numerous variables can affect the onset and course of pneumonia, making it a complicated disease. Pneumonia data has high dimensionality and noise, which makes data analysis difficult. This work describes a way for finding hidden patterns in chest X-ray pictures of patients with pneumonia by utilizing the Random Forest algorithm in conjunction with several Fisher face Techniques, including DLDA and PDLDA. One well-known difficulty that might occur when utilizing LDA (Linear Discriminant Analysis) for data analysis is the small sample size (SSS) problem. This issue arises when there are more variables (features) in the data than there are samples, making it impossible to estimate the covariance matrix. To address the intrinsic difficulties of Small Sample Size (SSS) in chest X-ray datasets, two methods are used in the data space: Direct Linear Discriminant Analysis (DLDA) and Parameterized Direct Linear Discriminant Analysis (PDLDA). The study looks at how well DLDA and PDLDA work to reduce high-dimensional data and extract discriminative characteristics from chest X-ray analysis. After that, the chest X-ray pictures are categorized using the Random Forest classification method according to the characteristics that were retrieved.