A hybrid machine learning technique for complex non-stationary classification problems

Abstract

Classification problems in machine learning have a wide range of applications including but not limited to, medical imaging, drug discovery, geostatistics, biometric identification, language processing, etc. In general, machine learning algorithms used for classification work on static input data. i.e. the number of classes in the dataset usually are known a priori or remains constant. In contrast, for several real-life applications, the data are dynamic and non-stationary in nature. The number of target labels is not fixed and can increase in real time. This results in an impending need to develop new machine learning methods to address sequential learning for non-stationary data samples featuring learning parameters. In this project, a novel technique that is independent of the number of class constraints and can adapt to the introduction of new classes it will encounter is developed. The developed technique will enable the system to remodel by itself adapting to dynamic needs of non-stationary input data samples. To be more specific novel machine learning technique based on Extreme Learning Machine is developed. Application of the proposed technique on several benchmark datasets demonstrate that the proposed technique is superior in terms of accuracy and consistency.