There are a plethora of algorithms in data mining, machine learning and pattern recognition areas. It is very difficult for non-experts to select a particular algorithm. Hence, according to current application or task at hand, recommendation of appropriate classification algorithm for given new dataset is a very important and useful task. According to NO-FREE-LUNCH theorem, there is no best classifier for different classification problems. It is difficult to predict which learning algorithm will work best for a particular type of data and domain. A meta-learning method is presented to support selection of candidate learning algorithms. Meta learning tries to address the problem of algorithm selection by recommending promising classifiers based on meta-features. It uses a k-Nearest Neighbor algorithm to identify the datasets that are most similar to the one at hand. The performance of the candidate algorithms on those datasets is used to generate a recommendation to the user in the form of a ranking. The performance is assessed using a multi-criteria evaluation measure that takes accuracy, precision, recall and execution time into account. The evaluation methodology is general and can be adapted to other ranking problems.
Pratik Jain
Anushka Kher
Supriya Shetty
Datasets taken from UCI repositories