A Hybrid Clinical Data Predication Approach Using Modified PSO

Abstract

Enhancement of diagnostic predictive mechanisms in Clinical Decision Support Systems (CDSS) is performed through improving disease staging predictions, illness progression prediction, and reducing the number of features in high dimensional clinical data sets. All the works in this paper uses benchmark datasets from University of California Irvine machine learning repository for verification. Clinical datasets are obtained through various diagnostics procedures, and the data from different sources was transformed into a single format before the start of the analysis. Since there are many diagnostics procedures, the transformed data usually has a large number of features. Preliminary data cleaning procedures like duplicate removal, noise dismissal, and filling up of missing data are performed initially. Data reduction techniques are applied over this cleaned data. A dataset is mainly due to two main reasons: too many instances or too many attributes. Once the noisy and duplicate instances are removed, having too many instances usually generates an accurate classifier. However, having too many features is a negative aspect of data analytics because the dataset's unimportant features tend to bring down the classifier's accuracy that has produced using the dataset. Reducing the number of features to an optimal level to enhance the classifier's accuracy is the aim of feature reduction algorithms. Every disease has various stages of severity. In this paper, an effective DSS support system for the Clinical Data predication Approach was proposed based the combination of Hybrid technique using C4.5, decision tree with Particle Swarm Optimization (PSO) evaluated for various diseases.