Proposing PSO Based Algorithm for Classifying Breast Cancer Data Effectively | Original Article

ABSTRACT:

Digital Image Processing is processing of images that are digital in nature by a digital computer. Thresholding is the most commonly used intensity-based image segmentation technique which converts gray scale images into binary image. The performance of thresholding algorithms mainly depends on selection of threshold value. Various statistical properties such as maximum likelihood, moment, entropy and between class-variance has been utilized for selecting a proper threshold. This study takestwo objectives in account. The first to develop an efficient segmentation algorithm based on PSO and second to test proposed algorithm in classifying risk thereby classifying breast cancer data more effectively and efficiently. The proposed hybrid approach for data mining has included two phases. In the ?rst phase, we adopted the statistical method in pre- processing. It can eliminate the insigni?cant features in order to reduce the complexity for next data mining stage. In the second phase, we proposed the data mining methodology that based on the standard PSO which called discrete PSO. In this study, we have used the Wisconsin breast cancer data set to test our proposed DPSO algorithm. In this study, a new hybrid approach of using both integrated statistical method and DPSO is proposed and successfully applied to the classi?cation risk of Wisconsin- breast-cancer data set.  According to our testing results, the proposed hybrid approach can improve the accuracy to 96.25%, sensitivity to 100% and speci?city to 96.32%.  These results are   very promising compared to the previously reported classi?cation techniques for mining breast cancer data.