Tutorial: Artificial Neural Networks to Analyze Single-Case Experimental Designs

Abstract

Since the start of the 21st century, few advances have had as far reaching consequences in science as the widespread adoption of artificial neural networks in fields as diverse as fundamental physics, clinical medicine, and social networking. In behavior analysis, one promising area for the adoption of neural networks involves the analysis of single-case graphs. However, few behavior analysts have any training on the use of these methods, which may limit progress in this area. The purpose of our tutorial is to address this issue by providing a step-by-step description on using artificial neural networks to improve the analysis of single-case graphs. To this end, we trained a new model using simulated data to analyze multiple baseline graphs and compared its outcomes to those of visual analysis on a previously published dataset. In addition to showing that artificial neural networks may outperform visual analysis, the tutorial provides information to facilitate the replication and extension of this line of work to other datasets and designs.

An Open-Source Relational Network Derivation Script in R for Modeling and Visualizing Complex Behavior for Scientists and Practitioners

“…Some of the computational analogs informed by RFT have been drawn from machine learning expert systems and artificial intelligence neural networks. Examples include the use of Deep Neural Networks (DNN) to detect effects in multiple baseline single case design graphed data ( Lanovaz and Bailey, 2020 ) and forecast human participant learning of trigonometry ( Ninness et al, 2019 ; Ninness and Ninness, 2020 ); the use of Kohonen Self-Organizing Maps (SOM: Kohonen, 1988 ) for behavioral pattern detection in legislature voting, breast cancer diagnosis ( Ninness et al, 2012 ), visual symmetry detection ( Dresp-Langley and Wandeto, 2021 ), and surgical expertise detection ( Dresp-Langley et al, 2021 ); and blends of DNN and SOM architectures to model decision making in child welfare systems ( Ninness et al, 2021 ). Additional work with Connectionist Models (CM) has provided confirmatory validation of methodological nuances in relational training sequencing for humans ( Lyddy and Barnes-Holmes, 2007 ).…”