Development of Multistage RFE-SVR Model to Predict Radiation Sensitivity

As radiation therapy (RT) has been in usage as a universal part of cancer patient treatment, the correlation between RT and patient profiles have become of interest to the research scholars and clinicians. There have been many studies which suggest a strong relation between radiation therapy and the genomic expression profiles of the patients. The analysis of the gene expression profiles poses a huge challenge due to the high dimensionality and the class imbalance problem. Identifying the best useful genes and eliminating the redundant ones is one of the key factors when analyzing genomic data. In our study, we have established a prediction model which identifies the radiation sensitivity index for the patients. For our study we have used SF2 (surviving fraction following irradiation with 2 GY) as the radiation sensitivity index. This multistage RFE-SVR model identifies the most useful features to predicts the SF2 value. We eliminate features step-by-step from the original set of features with a combination of filter and embedded method to avoid the loss of function.