Design of Interactive Feature Space Construction Protocol

Abstract

Machine learning deals with designing systems that learn from data i.e. automatically improve with experience. Systems gain experience by detecting patterns or regularities and using them for making predictions. These predictions are based on the properties that the system learns from the data. Thus when we say a machine learns, it means it has changed in a way that allows it to perform more efficiently than before. Machine learning is emerging as an important technology for solving a number of applications involving natural language processing applications, medical diagnosis, game playing or financial applications. Wide variety of machine learning approaches have been developed and used for a number of applications. newlineWe first review the work done in the field of machine learning and analyze various concepts about machine learning that are applicable to the work presented in this thesis. Next we examine active machine learning for pipelining of an important natural language application i.e. information extraction, in which the task of prediction is carried out in different stages and the output of each stage serves as an input to the next stage. newlineA number of machine learning algorithms have been developed for different applications. However no single machine learning algorithm can be used appropriately for all learning problems. It is not possible to create a general learner for all problems because there are varied types of real world datasets that cannot be handled by a single learner. For this purpose an evaluation of the machine learning algorithms is needed. We present an experiment for the evaluation of various state-of-the-art machine learning algorithms using an interactive machine learning tool called WEKA (Waikato Environment for Knowledge Analysis). Evaluation is carried out with the purpose of finding an optimal solution for a real world learning problem- credit approval used in banks. It is a classification problem. newlineFinally, we present an approach of combining various learners with the aim of increasing their efficiency. We present two experiments that evaluate the machine learning algorithms for efficiency and compare their performance with the new combined approach, for the same classification problem. Later we show the effects of feature selection on the efficiency of our combined approach as well as on other machine learning techniques. The aim of this work is to analyze the techniques that increase the efficiency of the learners.