LOGISTICS KERNEL PARTIAL Least Squares (KL-PLS) FOR REDUCTION nonlinear DIMENSIONS AND CLASSIFICATION BINARY (Case Study: Ovarian Cancer Data SELDI-TOF High Resolution)
ABSTRACT: Ovarian cancer is a cancer that is growing very rapidly. Development of early-stage cancer to an advanced stage can occur within a period of one year. Ovarian cancer is the fifth leading cause of death in Indonesia. Growth in the number of cancer patients is growing every year. For that, we need early diagnosis of cancer are needed to reduce the risk of death. A study conducted by Tenenhauset al. (2007) proposed a model that utilizes Logistics Kernel Partial Least Squares (KLPLS) to predict the presence of cancer. This thesis aims to utilize KLPLS method for predicting ovarian cancer using mass spectrometry data is SELDI-TOF ovarian High Resolution as proposed by Tang et al. (2010). KLPLS it self a development of the Partial Least Square involving the kernel function in the process. Kernel function used in the implementation of this method is a polynomial kernel function. KLPLS divided into two stages. The first stage is the stage trained to look for the regression coefficients. The second stage is the stage of the test to predict the test data labels based on regression coefficient of practice stage. The test is done to predict cancer in four different datasets. Based on trials KLPLS average yield of 93.53% best accuracy, sensitivity of 89.81% and a specificity of 98.89% in Data Set A. Based on the performance test data classification, it can be deduced that the classification model that utilizes KLPLS able to perform classification of ovarian cancer at the protein expression data SELDI-TOF format.
