Simulado AWS AI Practitioner AIF-C01

Recommending items based on a user's history and the behavior of similar users is the canonical recommendation (personalization) use case. Predicting an exact temperature is a regression problem because the target is a continuous numeric value, not a personalized item suggestion. Sorting tickets into predefined categories is a classification problem, where the output is a discrete label rather than a ranked recommendation. Detecting fraudulent versus legitimate transactions is also a classification (specifically anomaly/fraud detection) task, not a recommendation, since the goal is to assign a category, not to suggest items tailored to a user.

Each record carries a known label (churned or not), and the goal is to predict that label for new inputs, which is the defining characteristic of supervised learning. Unsupervised learning is wrong because it works on unlabeled data to find structure, but here labels exist. Reinforcement learning is wrong because there is no agent interacting with an environment to collect rewards; the model simply learns from a fixed labeled dataset. Semi-supervised learning is wrong (and self-contradictory) because the dataset is fully labeled, not unlabeled.

Regression predicts a continuous numeric value, which fits the dollar-amount prediction, while classification predicts a discrete category, which fits the spam-versus-not-spam decision. Swapping them is wrong because a dollar amount is continuous (not a category) and spam/not-spam is a discrete label (not a continuous number). Using regression for both is wrong because spam detection has discrete outcomes, not a numeric range. Using classification for both is wrong because predicting an exact dollar amount is a continuous-value problem, not a categorical one.