False positives in fraud detection AI systems often stem from poorly optimized hyperparameters. Hyperparameters control aspects of the model’s learning process such as the learning rate, decision thresholds, and complexity penalties. When these parameters are not tuned correctly, the model can become overly sensitive and flag normal behavior as suspicious, leading to customer complaints.

“Hyperparameter tuning is essential to balance sensitivity and specificity in AI models. Improper tuning can result in a high rate of false positives or negatives, particularly in systems like fraud detection that require nuanced pattern recognition.”

Options A and B relate to data governance but do not directly cause false positives in predictions. Option C (compute scale training) may affect model efficiency, not accuracy. Thus, D is the most appropriate answer.

In predictive maintenance use cases—such as detecting turbine failure—the most critical concern is identifying as many actual failures as possible to prevent catastrophic events. The AAIA™ Study Guide emphasizes that in such high-risk scenarios, Recall is the most appropriate metric because it measures the proportion of true positives correctly identified.

“Recall is critical in scenarios where missing a positive instance (e.g., a failure) is costly or dangerous. It ensures that most real issues are caught by the model, even at the expense of some false positives.”

Precision measures correctness of positive predictions, specificity measures true negatives, and accuracy may be misleading if the data is imbalanced. Thus, D (Recall) is most appropriate.

Human-centered design (HCD) focuses on integrating stakeholder needs, ethical implications, and social impact into AI system development. The AAIA™ Study Guide emphasizes the role of HCD in minimizing negative workforce impacts and ensuring inclusive design.

“Auditors should review whether the AI system design process included human-centered practices—particularly for applications affecting jobs or critical human roles. HCD ensures responsible adoption.”

While feedback collection (C) and impact assessments (D) provide useful context, A directly addresses ethical impact mitigation at the design level.