Meta-Learning in Supervised Machine Learning

Abstract

In the present digital era, a popular use of Machine learning is knowledge mining from big data. Machine learning is the sub-branch of Artificial Intelligence (AI) that extracts rules automatically from Big Data for decision-making to build expert systems. Meta-Learning is a sub-branch of machine learning, which uses machine learning classifiers that learns to map and combine predictions and information of data of other ML-models in the field of ensemble-learning. Meta-learning helps us to select the best/right learning algorithms to solve a particular problem. It maps from the meta-data of other machine learning algorithms by evaluating it on different datasets. In this paper, we have presented very recent state-of-the-art research works on meta-learning. We have categorized meta-learning on supervised learning data sets into three categories: (1) Task Independent Recommendation, (2) Configuration Space Design, and (3) Configuration Transfer, and reviewed the recent works on each category.