Limited access Past event
Applied Microeconomics Group Seminar What Does a Good Teacher Sound Like? Using Audio Files to Predict Teacher Effectiveness
Speaker: Professor Phil Oreopoulos, University of Toronto
Abstract: This project proposes a novel yet convenient method to predict a teacher's value-added. Using recent machine learning and deep learning techniques, we analyze audio files of teachers during recorded sessions and identify which teaching practices are most successful are increasing student test scores according to characteristics such as tone, pitch, and frequency of feedback. We also compare this machine learning approach with using humans to predict who is a very high versus very low value added teacher among randomly selected pairs. We hope to investigate whether data analysis by computers may improve predictive power of teacher value added and consider why this may be the case.