Nelson Baloian

Improving Accuracy and Interpretability of Regressors Using a Depster Shafer Classifier

Bio:

Dr. Nelson Baloian is Computer Engineering and PhD of Natural sciences. His research topics have been Computer Supported Collaborative Learning, Distributed Systems, GeoCollaborative System and AI for Decision Making. He has co-authored about 40 journal publications related on these subjects and more than 120 papers for peer reviewed conferences. He has been visiting professor at the universities of Waseda, Japan, and Duisburg-Essen, Germany. He is currently Associate Professor at the Department of Computer Science of the Universidad de Chile in Santiago, Chile, and program chair for the undergraduate programs of the department.

Description of the Talk:

Developing new regression techniques is a constant objective in machine learning research. The principal measures for evaluating the quality of a regression model are its accuracy and interpretability. However, experience has shown that accuracy and interpretability are often opposed goals; highly accurate regression methods often have low interpretability and vice versa. In this work, we present a methodology that combines the features of inherently interpretable methods with high-performing methods’ accuracy. We obtained a general regression method with high interpretability and precision simultaneously. In most cases, this method achieves higher accuracy than competing regression algorithms. The proposed model’s interpretability was validated by comparing it with a decision tree model, obtaining similar results.