Evaluating Pattern Classification Techniques of Neural Network Using k-Means Clustering Algorithm

Abstract

In the era of digitization, there is huge amount of digital data being processed and collected in the repositories. Lots of useful information and data patterns are hidden in this bulk data usually known as Big Data these days. So, it is now becoming important to store and manage this huge data for extracting important patterns and information for future decision-making. Classification is one of the important techniques while dealing with this huge amount of data. It is important to understand the diversity in the given set of data. Classification is the prediction of certain outcome on the basis of given input. In real life also, classification is the most common activity of human life. It is quite common phenomena of the day-to-day life, especially when we are involved in analytical task. It supports the decision-making task in business, research, etc. The classification problem is applicable on assigning label to an object among predefined group of elements on the basis of its properties and behaviour. Classification is the process of classifying the data in the different labels as per similarity measures of the defined groups. These days almost every field of research, medicine, business and industry, etc., are dealing with classification problems. Fraud detection, diagnosis of diseases, pattern recognition, loan approval and others are some of the examples of the classification problem. Data classification and clustering are the important techniques used in data mining. With the emergence of large volume of digital data, it becomes major challenge to manage this data in effective and efficient manner. Data mining is one of the techniques used for extracting required information from the large set of data. Various methods have been adopted in data mining in which data classification and clustering are used for labelling/grouping of data. Data classification may be defined as grouping of objects as per their similar characteristics on the basis of prior knowledge available to the system, i.e. classification groups the entities as per their similar features with available prior knowledge (supervised). Clustering is the method of grouping of objects as per their similar features without having prior knowledge (unsupervised). Neural network or artificial neural network is the mimic of human brain. It is the network of interconnected artificial nodes to process the information and provide its final output. Neural network is one of the tools of classification used in soft computing techniques. Various research studies have been done on neural network for classification task due to its best performance. Neural networks are treated as black box due to its hidden data processing. It can adjust its weight by itself. Usually, neural networks have the ability to learn the pattern itself by means of training process in the network. The error updation feature in the neural network adjusts the error after each iteration to make it more accurate (learning process of the ANN). Neural networks can approximate given function with arbitrary accuracy. Neural networks are nonlinear models, which help them flexible in modelling real world complex relationships. The objective of this paper is to evaluate pattern classification techniques of neural network using clustering technique. The focus of this paper is on the application of data mining in large database system and analysis of pattern classification techniques of neural networks in the database using k-means algorithm.