Meta-Learning for Escherichia Coli Bacteria Patterns Classification

In machine learning area, there has been a great interest during the past decade to the theory of combining machine learning algorithms. The approaches proposed and implemented become increasingly interesting at the moment when many challenging real-world problems remain difficult to solve, especially those characterized by imbalanced data. Learning with imbalanced datasets is problematic, since the uneven distribution of data influences the behavior of the majority of machine learning algorithms, which often lead to poor performance. It is within this type of data that our study is placed. In this paper, we investigate a meta-learning approach for classifying proteins into their various cellular locations based on their amino acid sequences, A meta-learner system based on k-Nearest Neighbors (k- NN) algorithm as base-classifier, since it has shown good performance in this context as individual classifier and DECORATE as meta-classifier using cross-validation tests for classifying Escherichia Coli bacteria proteins from the amino acid sequence information is evaluated. The paper reports also a comparison against a Decision Tree induction as base- classifier. The experimental results show that the k-NN-based meta-learning model is more efficient than the Decision Tree-based model and the individual k-NN classifier.