Abstract: Prediction influences the technological advancement in various sectors includes in finance for predicting the behavior of the stock market, in sports for predicting the outcome of the event, in opinion polls for predicting the outcome of the election and in many applications predicting their related unseen or unknown data. The prediction can be performed by the supervised learners also known as classifiers and their performance mainly relay on the variables taken to learning for building the predictive model hence, the variable should be relevant to the target concept of the leering process and these variables should not be redundant. Identifying the relevancy and redundancy of variables is called as variable selection. This is a preprocessing stage of knowledge discovery in prediction. Most of the variable selection processes are performed by some statistical or mathematical measures. This study presents a novel way of selecting the variables form the training dataset using unsupervised learners for enhancing the performance of the supervised learners in terms of increasing accuracy and reduce time taken to build the predictive model. The performance of this algorithm is evaluated with fourteen dataset with predictors namely Naive Bayes (NB), Instance Based (IB1) and tree based J48.
D. Asir Antony Gnana Singh, S. Appavu Alias Balamurugan and E. Jebamalar Leavline, 2014. Improving Performanc of Supervised Learners Using Unsupervised Variable Selection Algorithm: A Novel Approach. International Journal of Soft Computing, 9: 303-307.