IBM A1000-118 Advanced Practice Exam: Hard Questions 2025

A bank deploys a fraud-detection model and notices a sharp increase in false positives only for international transactions during holiday weekends. Labels arrive with a 2-week delay, and the feature set includes merchant category, amount, device fingerprint, and geolocation-derived distance. The team must quickly determine whether the issue is concept drift, data quality, or label leakage—without waiting for new labels. Which approach is most appropriate?

A healthcare provider builds a risk model for hospital readmission. They discover that the model’s predicted probabilities are well-ranked (good AUC) but systematically overestimate risk for low-risk patients and underestimate risk for high-risk patients. The model will be used to allocate limited care-management resources based on a probability threshold. What is the best next step to address this issue while preserving ranking performance?

A retailer trains a churn model on historical customer data. During validation, performance is excellent, but in production it collapses. Investigation reveals that a feature called "days_until_contract_end" is derived from a field that is updated after a customer calls to cancel (which happens near churn). This feature was available in training data but is not reliably available at scoring time for active customers. What is the most correct characterization and remediation?

A manufacturing company trains a model to predict equipment failure. Failures are rare (0.3%), and the cost of a false negative is much higher than a false positive. They currently report accuracy and are satisfied, but the maintenance team says the model misses critical failures. Which evaluation and operating-point strategy best aligns the model to the business objective?

A data science team uses k-fold cross-validation for a time-dependent demand-forecasting problem and observes very optimistic results. In production, the model performs poorly, especially around promotions. Which change most directly fixes the validation design flaw while preserving a rigorous estimate of generalization?

A company deploys a gradient-boosted model for loan approvals. Under new regulations, they must provide reasons for adverse actions to customers. The model uses hundreds of engineered features, including interaction terms. They need explanations that are consistent at the individual decision level and can be audited. Which approach is the best fit?

A support organization wants to implement a virtual agent that can answer FAQs and also perform account actions (e.g., reset password, check order status). They must minimize the risk of the assistant taking incorrect actions when user intent is ambiguous, and they need an escalation path to human agents. Which architecture is most appropriate using IBM Watson capabilities?

A team builds a knowledge-search experience over internal PDFs, policies, and web pages. Users complain that the system returns plausible but incorrect answers and cannot cite sources. The team must improve trust by grounding responses in enterprise content and showing verifiable references. Which solution best addresses the problem using IBM Watson technologies?

A regulated enterprise deploys an AI model and must demonstrate governance: lineage of training data, approval workflows, monitoring for drift, and the ability to reproduce a past prediction after an audit request. Which set of practices best meets these requirements end-to-end?

A company uses an AI system to screen job applicants. They find that a protected group has a substantially lower selection rate. Legal asks for a mitigation that improves fairness while maintaining job-relevant performance and providing defensible documentation. Which action is the most appropriate first step?