Talk 4
XI Workshop of Probability and Statistics group
CIDMA- University of Aveiro
Speaker: Luís Silva - DMat & CIDMA, University of Aveiro
Title of the talk: Predicting employee attrition using boosting models
Abstract: One of the big focuses of I4.0 is the creation of intelligent solutions for existing problems in industry. Anticipating the loss of qualified personnel, for example, is highly valuable as it can avoid needless costs. Employee attrition, the leave of a collaborator from the company, is a major concern for companies as losing highly qualified personnel has tremendous impacts in several aspects of its daily life. In this talk we discuss how machine learning tools can be used to tackle employee attrition using data collected from 3 companies pertaining to different Portuguese industrial sectors. We will show that boosting models, an ensemble technique where the result is the committee of many weak learners, are particularly useful for the problem at hand. The weak learners (usually decision trees with few nodes) are trained in sequence in a way such that observations wrongly predicted by a given weak learner get more weight/importance in the learning of the next one. In this way, and although they present an overall low performance, each weak learner specializes in a small subsample of the training data to compensate for the weakness of the previous one, resulting in a final strong ensemble learner. This work was conducted in the scope of the project Augmanity (www.augmanity.pt).Luís Silva, is graduated in Mathematics and Master in Statistics from the University of Porto. He received is PhD in the area of Machine Learning also from the University of Porto where he studied error-density risk functionals to train neural networks. He has been collaborating with several university and polytechnic institutions in the last 20 years. He is currently Assistant Professor at the Department of Mathematics of the University of Aveiro and collaborator of CIDMA-UA, in the Probability and Statistics Group and in the Biomath thematic line.
ORCID: 0000-0001-9677-4315