Investigating the Performance of Artificial Neural Networks in Predicting Affective Responses

Abstract

In this study it is aimed to examine the performance of an artificial neural network trained using items reflecting a latent trait in predicting responses to an item reflecting the same trait. This latent trait is the awareness of being able to communicate with people from different cultures, which is included in the PISA 2018 assessment. Relevant scale items were used as research variables. In addition to determining the extent to which the predicted responses overlap with the actual responses by analyzing the artificial neural network models, it was examined how the predicted responses affect the assumed latent construct and the reliability of the responses. Thus, the performance of artificial neural networks in predicting responses to affective items was evaluated. The responses expected from individuals for the items examined overlap with the responses given by individuals at a relatively moderate. However, it is observed that although the prediction values partially weaken the model fit indices, they still manage to keep them strong. In addition, the predicted values improved the factor loadings and the variance explained for the latent trait. Similarly, it is noticed that the predicted values also positively affect the reliability.