ROBUST CLASSIFICATION OF NOISE DATASET USING TYPE 2 FUZZY LOGIC SYSTEMS

Abstract

Classification is one of the very major areas in machine learning and image processing. Lot of techniques has been developed for classification but many fails with the presence of noise. The presence of noise in a measurement dataset has a negative effect on the classification model built. More specifically, the noisy instances in the dataset can adversely affect the learnt hypothesis. It is a difficult task to avoid 100% noise in measurement data while using a classifier. An efficient classifier should therefore be immune to noise and uncertain conditions. This dissertation presents an Interval Type 2 Fuzzy Logic (IT2 FL) based classifier for classifying noisy datasets. Fisher Iris and Wine dataset from UCI Repository are taken as examples. In machine learning, decision making, prediction and control system, it's important to have a very fast algorithm to calculate the output. Unfortunately fuzzy logic systems are computationally very intensive and interval type 2 fuzzy logic calculations are even difficult and time intense. This dissertation also presents a novel parallel algorithm for processing interval type 2 fuzzy logic calculations faster. The developed algorithm has been implemented in NVIDIA GeForce 9400M Graphics Processing Unit (GPU) and tested. The results shows that interval type 2 fuzzy logic system is very. effective in classifying noisy data sets and the response of the system is very fast while running in GPU.