Fundamentals of AI and ML

A retail analytics team has 12 months of accumulated sales records stored in Amazon S3. Once every night, they want to score the entire dataset with a trained model to generate a report for the next business day. There is no requirement for immediate, per-request responses, and no application waits synchronously for individual predictions. Which inference approach best fits this need?

A retail company deploys a machine learning model that classifies incoming customer support tickets into categories like 'Billing', 'Shipping', and 'Technical'. For each ticket, the model returns the predicted category along with a numeric value between 0 and 1 for each possible category. The support team wants to automatically route tickets only when the model is highly certain, and send low-certainty tickets to a human agent. Which characteristic of the model output should the team use to make this routing decision?

A telecom company has already built and validated a machine learning model that predicts whether a customer is likely to cancel their subscription. Each night, a batch job runs the current customer records through this model to generate a churn-risk score for every account, which the retention team uses the next morning. Which phase of the ML process does this nightly batch job represent?

A data scientist at a logistics company is building a machine learning model to estimate package delivery times. Before training, she divides her historical dataset into three separate portions. She uses one portion to fit the model, a second portion to tune hyperparameters and compare model configurations during development, and a third portion that she holds back and uses only once, at the very end, to report an unbiased estimate of how the model will perform on new data. What is the third, held-back portion of data commonly called?

A team is building an image recognition system that automatically identifies defective parts on a manufacturing line. Their approach uses a model with many layers of interconnected nodes that automatically learn hierarchical features (edges, then shapes, then whole objects) directly from raw pixel data, without engineers manually specifying which features to extract. Which subfield of AI/ML does this approach best describe?

A retail company has a dataset of 50,000 past customer support emails, each already manually tagged by staff as either 'billing', 'shipping', or 'technical'. The company wants to build a model that automatically assigns one of these three categories to each new incoming email. Which type of machine learning approach best fits this task?

A payments company is building a fraud detection model. Fraudulent transactions make up only 0.5% of all transactions in the training dataset. During evaluation, the team notices the model reports 99.5% accuracy but almost never flags real fraud. What is the most likely explanation for this outcome?

A marketing team wants to automatically produce unique first-draft product descriptions and promotional blog posts based on short prompts describing each new product. They want the system to generate original, human-like text rather than sorting existing content into categories. Which type of AI capability is most appropriate for this task?

A municipal records office receives thousands of handwritten paper forms each week. Staff currently type the contents into a database manually. The handwriting varies widely between individuals, and management wants to reduce the manual effort. Which factor most strongly indicates that a machine learning approach is appropriate for this problem?

A logistics company has trained an ML model that classifies package damage from photos. The model is now deployed, and every time a warehouse worker uploads a new photo, the system returns a damage/no-damage prediction within a second. In ML terminology, what is happening when the model generates a prediction for each newly uploaded photo?

A data science team at a logistics company is preparing a dataset to train a model that predicts package delivery delays. During review, they find that 40% of the historical records are missing the actual delivery timestamp, which is the value the model must learn to predict. Which statement best describes the impact of this issue on the training process?

A regional grocery chain wants to predict daily product demand for each store. Demand depends on many interacting factors such as seasonality, local weather, promotions, and holidays, and these patterns shift over time. The company currently uses a fixed spreadsheet formula that consistently over- or under-stocks. Why is a machine learning approach appropriate for this problem?

A finance team wants to automate the calculation of employee tax withholding. The rules are published by the government each year, are fully documented, and produce an exact, legally required result for any given salary and filing status. A developer suggests training a machine learning model on historical payroll records to predict the withholding amount. What is the best assessment of this proposal?

A retail analytics team is starting a new project to predict which products will sell out during seasonal promotions. They have identified the business goal and success metrics. Their next step is to gather three years of point-of-sale records, promotion calendars, and inventory logs from different databases, then clean inconsistent formats and handle missing values before any modeling begins. Which phase of the ML development lifecycle does this work primarily represent?

A data science team at a lending company has collected raw loan application data that includes an applicant's date of birth and total annual income. Before training a model to predict default risk, an analyst derives a new 'age' column from the date of birth and normalizes income into a scaled value. Which phase of the ML development lifecycle do these transformations represent?

A data science team has finished training a loan-approval classification model. Before they release it to the production endpoint, they want to measure how well the model generalizes to data it has never seen, using a portion of data that was held out from training. Which phase of the ML development lifecycle does this activity represent?

A retail company deployed an ML model that predicts weekly product demand. After six months in production, the operations team notices the model's forecasts are becoming increasingly inaccurate, even though the model code has not changed. Customer buying patterns have shifted due to new competitors entering the market. Which phase of the ML development lifecycle should the team focus on to address this problem?

A data science team trains a model to predict loan defaults. During evaluation, the model achieves 99% accuracy on the training data but only 62% accuracy on the held-out test data. What is the most likely explanation for this result?

A manufacturing company wants to use ML to predict the number of days remaining before a factory machine is likely to fail, so maintenance can be scheduled in advance. The output they need is a continuous numeric value (days until failure). Which type of ML problem best fits this requirement?

An e-commerce company wants to add personalized product recommendations to its website based on each customer's browsing and purchase history. The team has limited machine learning expertise and wants a managed AWS service that is purpose-built for generating real-time recommendations without requiring them to build and train a custom model from scratch. Which AWS service best fits this requirement?