A Novel Hybrid Nature Inspired Optimization Model (H-ACO) for Effective Mining of Medical Data of Diabetic Subjects for Computation of Risk Factor

Abstract

Data mining is one of the active areas of research in recent times especially with the advent of concepts of big data which are concerned with dealing of real time data. Data mining plays a vital role in health care for retrieval of health records from any point of globe at any point of time and also to predict certain occurrences/non occurrences of medical conditions based on previous history and symptoms. A nature inspired optimization approach is proposed in this research paper to estimate probability density function used in data clustering for identify the risk factors in diabetic patients based on hybrid ant colony optimization algorithm. HACO interprets a dataset as a spectral graph, whose nodes are the samples and each sample are connected to its nearest neighbours in a given feature space. Based on the pdf values the nodes of the graph are weighted and the distance between the samples are used to compute the pdf values and the k-nearest neighbours. Once the k-SGP is defined, HACO finds the root sample at each maximum of pdf and propagates to one optimum cluster from each sample to its remaining samples. Based on the pdf the clustering effectiveness is defined and the proposed model computes the best value of k for a given data mining application. We validate our approach in the context of classification and detection of diabetic patients. First, we compare HACO with data clustering based on k-means and spectral graph partitioning. Second, we evaluate proposed HACO technique to find the optimum value of k.