Author : Dhanashree Patil and Prof. Dhanashree Kulkarni

Biological field of research have already produced big amounts of valuable biological data that is challenging to analyze due to its high dimensionality and complexity. With this huge amount of data there are several problems during classification of data such as: over fitting, noisy instances and class imbalance. This will affect both the accuracy and the efficiency of supervised learning methods. This paper proposes a data-adaptive rulebased classification system for biological big data classification that generates relevant rules by finding adaptive partitions. The proposed system is a rule based classifier, which is combination of random subspace and boosting approaches. To construct the classification rules without global optimization, system makes use of J48 Decision Tree and KNN algorithm. Random subspace is used to avoid over fitting problem, boosting approach is used for solving problem of noisy instances classification and finally J48 decision tree is deal with class imbalance problem. With J48 decision tree, rules are evaluated from training data and with KNN, misclassified instances are analyzed. The proposed approach will be compared with other rule-based and machine learning classifiers, and detailed results and discussion of the experiments are presented to demonstrate comparative analysis and the efficacy of the results. Results will prove the good prediction accuracy of classified DNA.

Keywords: Biological data, J48 classifier, KNN classifier, misclassified instances.