Dimension reduction: A PSO-PCNN optimization approach for attribute selection in high-dimensional medical database

Data mining is the process of analyzing data from various perspectives and summarizing into useful information. Data mining is applied to find the hidden patterns in the important roles of medical diagnosis and prognosis. Dimension reduction is an important topic in data mining, which is widely used in the areas of genetics, medicine and bioinformatics. The medical data are paradoxical and hundreds of independent features in them. In order to make the valuable decisions and to improve the efficiency, accuracy of mining tasks on high dimensional data, we propose a new dimension reduction (Feature selection & Extraction) algorithm a hybrid model of PSO-PCNN. From the proposed algorithm, the best attributes from the Particle Swarm Optimization (PSO) will be passed as input to the Pulse coupled Neural Network (PCNN) for further optimization. The datasets collected from the several Diabetic public health care centers, it was about 470 instances and also having 28 attributes of Impaired Glucose Tolerance (IGT) patient with Glucose Tolerance Test (GTT). This paper focuses on extracting the best potential attributes from the paradoxical high dimensional medical datasets. The proposed algorithm reduces these high dimensional features in to 13 attributes with limited number of iterations. Our comparisons are performed in terms of relevance and redundancy. The obtained experimental result shows very promising outcomes for the computational efficiency and also to improve the classification accuracy.