CHAPTER 3 CILOP: CLUSTERING BASED METHOD FOR CLASS IMBALANCE LEARNING USING OPTICS The present chapter proposes a novel approach that is Clustering based Class Imbalance Learning using OPTICS [49] method for improvement of class imbalance learning. The content of this chapter is published in “International Journal of Computer Applications”, Page No: 33-42, Volume 51-No 16, August 2012, K.Nageswara Rao , T.Venkateawara Rao and D.Rajya Lakshmi, “A Novel Class Imbalance Learning using Ordering Points Clustering”. The reprint is appended towards end of this thesis. 3.1 INTRODUCTION This chapter mainly concentrates on for developing practical algorithms which are deal with the class imbalance learning problems. We have proposed a new class imbalance learning method Class Imbalance Learning using OPTICS (CILOP). In CILOP, the clustering based approach is used to improve class imbalance learning using under-sampling. This under-sampling CIL approach overcomes the weakness of removing useful instances. OPTICS is one of the well-known clustering techniques. Its order based approach is one of the efficient frameworks for analyzing datasets of different size and properties, which are not easily to analyze. OPTICS generates density based clusters as DBSCAN with considering ordered approach for analysis. Figure 3.1 description of cluster. The advantage of OPTICS approach over other clustering approach is the unique way of analysis using visualization technique. Figure 3.1 depicts two types clusters one is weak cluster and other one is strong cluster. In weak cluster the objects are wider distance than in strong cluster because the properties of objects are dissimilar. But so far no researcher used OPTICS clusterin... ... middle of paper ... ... subset. The instances with the higher merits found will be retained in the majority subset for producing quality results. The experiments conducted with CILOP specify that improved CIL measures can be achieved. We have conducted experiments on 10 datasets from UCI which suggest that CILOP can quickly remove redundant, irrelevant and weak instances as long as the properties of the dataset are normal. Excellent improvement in CIL measures on some natural domain datasets shows the compatibility of CILOP approach on real-time applications. One of the shortcomings seen in CILOP is when used for datasets with unique properties. It is so because CILOP will not consider unique properties of datasets for removing instances from majority subset. Finally, we can conclude that CILOP can be a good contribution as a CIL method for efficient learning of the imbalanced datasets.