A comparison of classification methods in vertebral column disorder with the application of genetic algorithm and bagging

Disorders of the spine are experienced by about two-thirds of adults and belong to the second most common disease after headache. Prediction of spinal disorders is difficult because it requires an experienced radiologist to analyze images of Magnetic Resonance Imaging (MRI). The use of Computer Aided Diagnosis (CAD) system can help the radiologist detect abnormalities in the spine more optimal. In the vertebral column data set which is available now has three classes that indicate the condition of the spine, that are herniated disk class, spondylolisthesis class and normal class. As well as on the data sets that has several classes, there are problems called the class imbalance which causes a lack of accuracy in the classification results. In this study, the combination of genetic algorithm and bagging technique are proposed to improve the accuracy of class classification on spinal disorders. Genetic algorithm is used for feature selection while bagging technique is used to solve the problem of class imbalance. The proposed method is applied to three classifier algorithms, namely naïve bayes, neural networks and k-nearest neighbor. The results showed that the proposed method makes a significant improvement in the classification of disorders of the spine for most classifier algorithms. The best algorithm after applied to genetic algorithms and bagging technique is k-nearest neighbor with an accuracy of 89.03%, 88.06% for the neural network and 86.13% for naïve bayes if validated using cross validation.