IBM A1000-080: Assessment: Data Science and AI Practice Exam 2025: Latest Questions

A data scientist is asked to summarize a dataset that contains customer ages with a few extreme outliers (e.g., 5 and 120). Which measure is MOST appropriate to describe the typical age?

A team trains a classification model and obtains high accuracy, but the dataset is highly imbalanced (95% of records are class 0). Which metric is MOST appropriate to evaluate performance on the minority class?

A project team is moving from exploratory analysis to building a predictive model. Which step should be performed FIRST to reduce the risk of data leakage?

You need to quickly build a baseline model to predict a continuous target such as house price. Which algorithm is the MOST appropriate starting point?

A churn model performs well on the training set but significantly worse on the test set. Which action is MOST likely to reduce overfitting?

A dataset includes a categorical feature "city" with hundreds of possible values. You want to use it in a linear model. Which encoding approach is MOST appropriate in general?

A computer vision model recognizes products on a shelf. The team has a small labeled dataset and wants to improve performance quickly. What is the BEST approach?

A team wants to ensure their notebook-based experimentation can be reproduced and reviewed. Which practice is MOST appropriate?

A model is deployed to score loan applications. After deployment, the approval rate drops sharply even though the incoming population seems similar. Investigation shows the distribution of an input feature (income) has shifted due to a new data source. What should the team do FIRST?

A deep learning classifier achieves 99% training accuracy but only 70% validation accuracy. The team also notes validation loss increases after a certain number of epochs while training loss continues to decrease. Which technique is MOST appropriate to apply next?

A data scientist is exploring a new dataset and wants to understand whether two numerical variables move together linearly, including the direction and strength of that relationship. Which metric is most appropriate?

A team built a binary classifier to detect fraudulent transactions. Fraud is rare (about 1% of transactions). Accuracy is high, but many fraud cases are missed. Which evaluation metric should the team prioritize to reduce missed fraud cases?

In a deep learning workflow, what is the primary purpose of a validation dataset?

A retailer wants to segment customers into groups based on purchasing behavior without labeled outcomes. Which approach is most appropriate?

A model performs exceptionally well on training data but significantly worse on unseen data. Which action is most likely to improve generalization?

A data engineer notices that a dataset contains many missing values in a numeric column that is important for modeling. The missingness is not random (e.g., values are missing more often in one customer segment). What is the best practice before choosing an imputation strategy?

An organization wants a repeatable process to build, test, and deploy ML models with traceability from data to model artifacts. Which practice best supports this goal?

A project team is selecting a cloud service to discover, understand, and govern enterprise data assets using a shared metadata catalog and business glossary. Which IBM capability aligns best with this requirement?

A binary classifier is trained on historical loan approvals. During evaluation, the model shows a large difference in false negative rates between two demographic groups. What is the most appropriate next step?

A team builds a model to predict equipment failure. They randomly split sensor readings into train/test sets. The test performance is extremely high, but in production the model degrades quickly. Investigation shows that readings from the same machine and time period appear in both train and test sets. What is the most likely issue and best fix?

A data scientist is exploring a new dataset and discovers some columns have missing values. Before building any predictive model, which action is the best first step to determine an appropriate handling strategy?

A team evaluates a classifier on a dataset where only 1% of cases are positive (rare event detection). They report 99% accuracy, but the business says the model misses most positive cases. Which metric is most appropriate to emphasize to better reflect performance on the positive class?

A bank is building an AI solution using foundation models to assist customer support agents. They want to reduce the chance the system produces unsafe or policy-violating content. Which approach best addresses this requirement?

A team uses a scikit-learn pipeline that applies standardization and one-hot encoding before training a model. In production, predictions are significantly worse than in validation, even though the model artifact was deployed correctly. Which issue is the MOST likely cause?

A retail company wants to build a solution that answers questions using internal policy documents. The documents change frequently, and the company wants answers grounded in the latest content without retraining a large model each update. Which architecture is the best fit?