Intro to Machine Learning Foundations
Free Oracle Cloud Infrastructure 2025 AI Foundations Associate practice — 6 questions on Intro to Machine Learning Foundations, with explanations. No sign-up. Full 12-question mixed test →
A data science team trains a decision tree that achieves 99% accuracy on the training dataset but only 61% accuracy on a held-out test dataset. Which condition BEST explains this behavior?
A large gap between very high training accuracy and much lower test accuracy is the classic signature of overfitting: the model has fit noise/idiosyncrasies of the training set rather than the general pattern, so it fails to generalize to unseen data.
An engineering team wants an agent to learn how to balance a robot on two wheels. The agent receives a positive signal when the robot stays upright longer and a negative signal when it falls, with no labeled 'correct action' provided for each state. Which machine learning paradigm does this scenario describe?
The scenario describes an agent interacting with an environment, taking actions, and receiving reward/penalty signals rather than labeled input-output pairs. This trial-and-error, reward-driven learning process is the defining characteristic of reinforcement learning.
A retailer has a dataset of 50,000 customer transactions with no predefined categories or labels. The team wants to discover natural groupings of customers with similar purchasing behavior so they can design different marketing campaigns for each group. Which approach is MOST appropriate?
Since there are no predefined labels and the goal is to discover natural groupings based on similarity, this is a clustering problem, which falls under unsupervised learning.
A dataset for detecting a rare disease contains 990 healthy patients and only 10 patients with the disease. A model that simply predicts 'healthy' for every patient achieves 99% accuracy. Why is accuracy alone a poor conceptual choice for evaluating this model?
On highly imbalanced datasets, a naive model that always predicts the majority class can achieve high accuracy while providing zero value for detecting the minority (and often more important) class, which is why accuracy alone can be a misleading conceptual metric in such cases.
During an ML workflow, a data scientist takes raw transaction timestamps and converts them into new inputs such as 'day of week', 'hour of day', and 'time since last transaction' before training a fraud detection model. Which step of the machine learning workflow does this represent?
Transforming raw data (timestamps) into derived, more informative inputs (day of week, hour, time since last transaction) to improve model performance is the definition of feature engineering, a preparation step that occurs before training.
A team splits their labeled dataset into 80% for training and 20% for testing before building a classification model. What is the PRIMARY purpose of reserving the 20% test set rather than using all data for training?
Holding out a test set that the model never sees during training allows an unbiased evaluation of generalization performance on unseen data, which is the primary reason for the train/test split in the ML workflow.
Ready for the real thing?
The full course has two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed answer explanations.
Start my full course on Udemy →