Extracting Meta Knowledge from Machine Learning and Data Mining Algorithm

Abstract

Extracting meta-knowledge from real-world meta-data is a challenging task, which is a fundamental conceptual instrument for knowledge engineering and knowledge management. Meta-knowledge is a knowledge that learns to employ meta-learning in the field of machine learning and data mining. Recently, meta-information are using for decision making in many real-life machine learning for data mining applications. In this thesis, we have proposed a new method for extracting meta-knowledge applying machine learning algorithms. We have considered 10 most popular supervised learning algorithms (e.g. naive Bayes Classifier, Decision Tree, RandomForest, RandomTree, Bagging, Boosting, OneR, PART, Support Vector Machine, k-Nearest Neighbor) and 51 benchmark datasets from UC Irvine Machine Learning Repository. Initially, we have applied the learning algorithms on the datasets and engender the experimental results based on accuracy with 10-fold cross-validation and test data (30% split). Then, we have produced meta-data from the experimental results. To generate meta-data we added several extra features (i.e. size of the dataset, area, dimension, attribute types, is missing data present, streaming data etc.). Finally, we have applied decision tree induction algorithm on the meta-datasets to get meta-knowledge. We have tested our proposed model on several test data. The experimental results show that our proposed model achieved about 98% accuracy.