The Effectiveness of Machine Learning in Detecting Fakers Across Different Assessment Methods and Multiple Algorithms

Abstract: Faking, a prevalent phenomenon in noncognitive assessments, is a concern because of its potential negative effects on criterion-related validity. This study contributes to the faking detection literature by examining the effectiveness of machine learning (ML), specifically 8 algorithms, to classify fakers based on their responses on one of four assessment methods (single-statement Likert, single-statement true/false, a rate situational judgment test, and a rank situational judgment test). Using an MTurk sample ( N = 583), faking was manipulated in a between-subjects design. Classification accuracies comparable to those of previous studies were achieved for two of the four assessment methods albeit by different algorithms indicating that the effectiveness of the ML approach should not be blindly assumed and multiple algorithms should be considered.