Featureless Classification Model Training Algorithm Based on Similarity Measure

Abstract

For the learning problems of vectorial data many solutions and algorithms have been developed. But in physical world data is depicted as feature vectors. Domains like computer vision, bioinformatics the data is not available as vectorial data but as pair-wise data. The project proposes the new method for similarity based learning using distance transformation and distance is taken as the similarity measure. For a data set a clear class boundary is generated to identify the class by manipulating the distance between the data points. The proposed method is developed using two algorithms; Pair-wise Similarity Based Classifier which is used for train data set and Classifying Unknown Data Points used to label the class for unknown data point. The outcome of the proposed method is compared with k-nearest-neighbor classifier. The accuracy rate is increased and error rate is reduced in proposed method.