ATOM – a flexible multi-method machine learning framework for predicting job success

Gergely, Bence and Takács, Szabolcs (2023) ATOM – a flexible multi-method machine learning framework for predicting job success. APPLIED PSYCHOLOGY IN HUNGARY, 25 (3). pp. 17-32. ISSN 1586-7382

Preview

Text APA_2023_3_2.pdf Download (528kB) | Preview

Abstract

Background and Aims: Presenting the statistical fundamentals of ATOM and its concurrent algorithms, with particular respect to demonstrate the flexibility of the decision-making module. Methods: Simulating different classification problems using the Scikit Learn machine learning program package. During these simulations, the sample size, the number of variables, the number of groups, the proportion of incorrect classifications, and the distance between the groups were systematically changed. Results: Based on 180 datasets, the Multilayer Perceptron performed the best in about 52% of the cases, and the Support Vector Classifier came in second place. It was found that every method proved to be better than any other in at least one case, which means that if we are dealing with a company or job where the given problem arises, these procedures provide a more accurate result. In addition, profound differences between different parameters of the same procedure were observed. Discussion: Considering that the job selection aims to filter the best candidates, the accuracy of all procedures increases and, in general, it was shown that ATOM’s algorithms indicate a performance much above the expected value of random categorization.

Actions (login required)

Edit Item