Psychological Forest: Predicting Human Behavior

Ori Plonsky, Ido Erev, Tamir Hazan, Moshe Tennenholtz

Abstract: We introduce a synergetic approach incorporating psychological theories and machine learning in service of predicting human behavior. Our method harnesses fundamental psychological theories to provide rigorous features to a machine learning algorithm. We show that this approach can be extremely powerful. In particular, a random forest algorithm, dubbed psychological forest, which makes use of the psychological features in addition to the more standard objective and naive features, leads to prediction that significantly outperforms best practices in a choice prediction competition. Our results suggest that the main power of this synergy between psychological theories and machine learning is evident when facing a completely novel setting rather than completing missing values of a partially known matrix.