Lanovaz Marc J, Hernandez Varsovia, León Alejandro
École de psychoéducation, Université de Montréal, Canada.
Centre de recherche de l'Institut universitaire en santé mentale de Montréal, Canada.
J Exp Anal Behav. 2025 Jul;124(1):e70029. doi: 10.1002/jeab.70029.
Traditionally, the experimental analysis of behavior has relied on the single discrete response paradigm (e.g., key pecks, lever presses, screen clicks) to identify behavioral patterns. However, the development and availability of new technology allow researchers to move beyond this paradigm and use other features to detect schedules. Thus, our study used spatiotemporal data to compare the accuracy of four machine learning algorithms (i.e., logistic regression, support vector classifiers, random forests, and artificial neural networks) in detecting the presence and the components of time-based schedules in 12 rats involved in a behavioral experiment. Using spatiotemporal data, the algorithms accurately identified the presence or absence of programmed schedules and correctly differentiated between fixed- and variable-space schedules. That said, our analyses failed to identify an algorithm to discriminate fixed-time from variable-time schedules. Furthermore, none of the algorithms performed systematically better than the others. Our findings provide preliminary support for the utility of using spatiotemporal data with machine learning to detect stimulus schedules.
传统上，行为的实验分析依赖于单一离散反应范式（例如按键、压杆、点击屏幕）来识别行为模式。然而，新技术的发展和可用性使研究人员能够超越这一范式，利用其他特征来检测时间表。因此，我们的研究使用时空数据比较了四种机器学习算法（即逻辑回归、支持向量分类器、随机森林和人工神经网络）在检测参与行为实验的12只大鼠中基于时间的时间表的存在及其组成部分时的准确性。利用时空数据，这些算法准确地识别了编程时间表的存在与否，并正确区分了固定空间和可变空间时间表。也就是说，我们的分析未能找到一种算法来区分固定时间和可变时间时间表。此外，没有一种算法在系统上比其他算法表现得更好。我们的研究结果为使用时空数据和机器学习来检测刺激时间表的实用性提供了初步支持。
