A subscription box company wants to predict which of its current customers are likely to cancel their subscription in the next 60 days, using each customer's order history, support ticket count, and login frequency, so the retention team can reach out first. Which machine learning approach fits this need?
Select an answer to reveal the explanation.
Short Explanation
The retention team has a very specific question in mind for every customer: will this person cancel or not. That is a two-outcome prediction problem, and the machine learning task built for sorting things into known buckets like that is the one to reach for here. Grouping similar customers together sounds tempting, but grouping does not actually tell you which group is the cancel-soon group, it just clusters people who look alike, leaving you to figure out the labels yourself. Flagging weird or rare behavior is a different job entirely, aimed at catching outliers rather than predicting a common, expected outcome that happens to plenty of customers. And predicting a dollar figure like future spend is a numeric estimation task, which is a different shape of problem than predicting a category. Since the team wants a yes or no per customer, the labeled-category approach is the match.
Full Explanation
The correct answer is B. Classification is the right fit because the company needs to predict a discrete, labeled outcome for each customer, cancel or stay, based on input features like order history, support tickets, and login frequency, and classification models are trained specifically to assign records to known categories like this. Option A is incorrect because clustering finds natural groupings in data without using a target label, so it could reveal customer segments but would not directly predict whether a specific customer will cancel. Option C is incorrect because anomaly detection is aimed at spotting rare or unusual data points, such as fraudulent account access, rather than forecasting a common business outcome like churn across the whole customer base. Option D is incorrect because regression predicts a continuous numeric value, such as a dollar amount for lifetime spend, not a categorical yes or no outcome like cancellation.