A Prediction Model for PCOS by Using Enhanced Bio Inspired Harmony Search Based Wrapper Approach

Abstract

The analysis of health and medical data is crucial for improving the diagnosis precision, treatments and prevention.  Data mining, a field that can uncover patterns from large repositories, has numerous applications such as building predictive models which can be extremely beneficial in the healthcare industry.  Due to the development of modern technology, data mining applications in healthcare consist about the analysis of health care centres for enhancement of health policy-making and prevention of hospital errors, early detection, prevention of diseases and preventable hospital deaths, more value for money and cost savings, and detection of fraudulent insurance claims. Polycystic ovary syndrome (PCOS) is the most common endocrinopathy in women, primarily affecting the reproductive system, with substantial collateral negative health effects on metabolic,  psychologic,  and cardiovascular functions. Early detection and treatment of PCOS is important since it is largely prevalent among women of reproductive age. A recent study has revealed that about 18% of the women in India suffer from this syndrome. Even though PCOS has been identified as the most common endocrinal disease , the author has found that there are a very limited number of researches initiated towards building a prediction model for PCOS. Hence, the objective of this work  is to build an efficient  prediction model  to predict whether a person is likely to have PCOS or not by applying meta-heuristic based  Feature Selection  methods  to improve the accuracy of data classification for the  PCOS dataset so that the improved accuracy will better yield enough information to identify the potential patients and thereby improvise the diagnosis accuracy. In this work, in the attempt to design an efficient prediction model, three meta-heuristic algorithms such as Particle Swarm Optimization Algorithm(PSO), Genetic Algorithm(GA) and  Bio inspired Harmony Search Algorithm(BioHS) were used for feature selection and the efficiency of each method was analysed by using three various classifiers such as Support Vector Machine, Naive Bayes and K Nearest Neighbour.  The performance of all  combinations of the optimization method and the Classifiers on the PCOS dataset were analysed and it was found that the Bio inspired Harmony Search and  K Nearest Neighbour combination outperformed the other combinations.