Towards creating a reference based self-learning model for improving human machine interaction

Abstract

Machine learning is a field of computer science where a computer is provided a capability to learn rather being programmed every time. Machine learning is considered to be the next generation human machine interaction technology, where a machine will independently learn from users previous data and provide solution and better suggestions to the user. In this work a reference based self-learning model is proposed, which can learn classification on new data from its previous trained models. Here a classification problem is considered to have a series of events related to every class. For classification, all the events related feature vector of a class is fed to the model. Now instead of learning the classification for each event based feature separately, reference based learning is adopted, where the model is trained for only one event based feature and then it learns the classification for other event features by itself. A simulation is performed on three feature vectors as a process of events and the results are presented. The model achieves an accuracy of around 90% using reference based learning. The achieved result of reference based learning is encouraging as it is very close to separately trained model with increase in number of samples per class. This method of live training and classification reduces time required for database preparation and model training separately for each event based features.