This paper presents a sandbox example of how the integration of models borrowed from Behavioral Economic (specifically Protection-Motivation Theory) into ML algorithms (specifically Bayesian Networks) can improve the performance and interpretability of ML algorithms when applied to Behavioral Data. The integration of Behavioral Economics knowledge to define the architecture of the Bayesian Network increases the accuracy of the predictions in 11 percentage points. Moreover, it simplifies the training process, making unnecessary training computational efforts to identify the optimal structure of the Bayesian Network. Finally, it improves the explicability of the algorithm, avoiding illogical relations among variables that are not supported by previous behavioral cybersecurity literature. Although preliminary and limited to 0ne simple model trained with a small dataset, our results suggest that the integration of behavioral economics and complex ML models may open a promising strategy to improve the predictive power, training costs and explicability of complex ML models. This integration will contribute to solve the scientific issue of ML exhaustion problem and to create a new ML technology with relevant scientific, technological and market implications.
翻译:本文展示了一个沙箱实例,说明从行为经济学(特别是保护-动力理论)中借用的模型与ML算法(特别是巴伊西亚网络)的整合如何在应用行为数据时改进ML算法的性能和可解释性。行为经济学知识整合以界定巴伊西亚网络的架构,提高了11个百分点的预测的准确性。此外,它简化了培训过程,为确定巴伊西亚网络的最佳结构进行了不必要的培训计算工作。最后,它改进了算法的可复制性,避免了以往行为网络安全学文献不支持的变量之间的不合逻辑关系。尽管我们的结果表明,行为经济学和复杂的ML模型的整合初步和限于零ne简单模型,可以开启一个有希望的战略,改进复杂的ML模型的预测力、培训成本和可复制性。这种整合将有助于解决ML耗竭的科学问题,并产生具有相关科学、技术和市场影响的新的ML技术。