A HYBRID SELF ORGANIZING MAP IMPUTATION (SOMI) WITH NAÏVE BAYES FOR IMPUTATION MISSING DATA CLASSIFICATION

Abstract

This study proposes hybrid SOMI (Self Organizing Map Imputation) and Naive Bayes (NB) model on data, that contain missing values to improve the performance of the Naive Bayes Imputation (NBI) it has weaknesses for missing categories n ≤ 1. This new hybrid model, using imputation approach based on SOMI is used for prepossessing and NB classification for the classification process in multivariate data, so that it can improve performance. SOMI measurements use an average error with self-organizing feature map. The multivariate attribute is converted to numeric attributes to establish data uniformity. The SOMI learning results have used weight variations by combining the mechanism of distance hierarchical value representation with a new scheme to overcome mixed types. Hybrid SOMINB is used to classify mixed data to correct misclassification. The model has advantages because it can update weights with the probability of each attribute. Attribute values have produced a set of probabilities for each cluster using the Naive Bayes group. Outputs of the SOMI Method are used as learning machines to produce training data for the target class to be used in Naive Bayes machine learning. The results of this study used all missing scenarios at a random mechanism and various missing percentages. The results of the hybrid SOMINB model showed more results with an accuracy rate of 90.00% with other imputation analysis. Experimental results present that the proposed produces higher accuracy than general estimating values which established missing value treatment methods.