Laith T. Khrais

Type of the article: Research Article

Abstract
The increasing complexity of insurance fraud in Jordan has unveiled inadequacies of traditional detection mechanisms, calling for advanced technologies. This study investigates drivers and inhibitors of machine learning adoption for fraud detection within Jordan’s insurance sector, with a focus on institutional readiness, ethical concerns, and supporting regulations. By applying quantitative and exploratory research design, Partial Least Squares Structural Equation Modeling serves as an approach to analyze data collected from 291 practitioners of fraud detection, data science, and insurance compliance in the industry.
Findings show that both existing fraud detection efforts (coefficient = 0.42, p = 0.012) and knowledge of machine learning (coefficient = 0.55, p = 0.009) have favorable impacts on adoption likelihood, which underlines the relevance of bureau experience and informed professional culture. By contrast, major adoption deterrents such as limited IT capability, budgetary constraints, and moral concerns about fairness and clarity (coefficient = –0.40 and –0.38, respectively) unfavorably decrease adoption intention.
Regulatory encouragement has a two-fold role: it has a direct promoting effect on adoption (coefficient = 0.47, p = 0.011) and a buffering effect on negative ethical concerns (interaction = 0.36, p = 0.025) and adoption barriers (interaction = –0.28, p = 0.032). Perceived efficacy also mediates between awareness/experience on the one hand and adoption decisions on the other (coefficients = 0.51 and 0.44, p < 0.05).
The results demonstrate successful incorporation of machine learning into fraud detection as depending on the clarity of regulations, ethical protections, and institutional readiness, rather than on technical capability itself.