Loading...
Paper Number
2139
Paper Type
Short
Description
Machine learning models are widely used in many business contexts, but there is a growing concern that strategic individuals may manipulate their features to obtain desirable outcomes from the machine learning models. This paper offers a theoretical analysis of the impact of feature manipulation on the performance of the machine learning models and the payoffs of firms in an online lending context. Contrary to the common belief, our interesting finding is that manipulation may not be harmful to a firm under some circumstances. Instead, it could increase the classification model's performance and raise a firm's payoff and the social welfare when high-quality individuals manipulate more. Overall, our findings suggest that manipulation can bring strategic value to machine learning models instead of just being a harmful activity. Our findings provide useful insights for feature engineering and lay a foundation for future research about optimal strategies to cope with manipulation activities.
Recommended Citation
ZHOU, Jiali and ZHENG, Jiexin, "A Strategic Analysis of Algorithm Manipulation: a Lending Game perspective" (2022). ICIS 2022 Proceedings. 15.
https://aisel.aisnet.org/icis2022/ai_business/ai_business/15
A Strategic Analysis of Algorithm Manipulation: a Lending Game perspective
Machine learning models are widely used in many business contexts, but there is a growing concern that strategic individuals may manipulate their features to obtain desirable outcomes from the machine learning models. This paper offers a theoretical analysis of the impact of feature manipulation on the performance of the machine learning models and the payoffs of firms in an online lending context. Contrary to the common belief, our interesting finding is that manipulation may not be harmful to a firm under some circumstances. Instead, it could increase the classification model's performance and raise a firm's payoff and the social welfare when high-quality individuals manipulate more. Overall, our findings suggest that manipulation can bring strategic value to machine learning models instead of just being a harmful activity. Our findings provide useful insights for feature engineering and lay a foundation for future research about optimal strategies to cope with manipulation activities.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
10-AI