A Survey & Current Research Challenges in Meta Learning Approaches based on Dataset Characteristics

Classification is a process that predicts class of objects whose class label is unknown. According to No Free Lunch (NFL) theorem, there is no single classifier that performs better on all datasets. Meta learning is one of the approaches that acquired knowledge based on the past experience. The knowledge in Meta-Learning is acquired from a set of meta-examples which stores the features of the problem and the performance obtained by executing a set of candidate algorithms on Meta Features. Based on the experience acquired by the system during training phase, ranking of the classifiers is provided based on considering various measures of classifiers.