Describe fundamental principles of machine learning on Azure

The correct answer is Label.

In supervised learning, a label is the target variable or known outcome that we want the machine learning model to predict. It represents the 'answer' or ground truth for each observation in the training dataset. For example, in a dataset used to predict house prices, the actual house price would be the label. In a spam detection system, whether an email is 'spam' or 'not spam' would be the label.

The training process in supervised learning works by showing the algorithm both the input features and their corresponding labels, allowing it to learn the relationship between them. Once trained, the model can then predict labels for new, unseen data.

Why the other options are incorrect:

Feature attribute refers to the input variables or characteristics used to make predictions, not the outcome itself. Features are the descriptive properties of each observation (like square footage, number of bedrooms for house price prediction), while the label is what we're trying to predict.

Training parameter refers to the internal values that the machine learning algorithm adjusts during the learning process (such as weights in a neural network). These are not part of the dataset but are learned by the model to minimize prediction errors.

Predictor variable is actually another term for features or independent variables - the inputs used to make predictions. While it might seem like it could refer to the outcome, in statistical and machine learning terminology, predictor variables are what you use to predict, not what you're predicting.

The correct answer is Unsupervised learning.

Clustering is a fundamental technique in unsupervised learning where the algorithm analyzes data to find natural groupings or patterns without being provided with labeled training data. In clustering, the machine learning model examines the features of data points and groups similar items together based on their characteristics, without any prior knowledge of what the groups should be or what they represent.

Key characteristics of clustering as unsupervised learning:
- No labeled data is required for training
- The algorithm discovers hidden patterns and structures in the data on its own
- Common clustering algorithms include K-Means, Hierarchical Clustering, and DBSCAN
- Used for tasks like customer segmentation, anomaly detection, and data exploration

Why the other options are incorrect:

Supervised learning with categorical outputs (also called classification) requires labeled training data where each example has a known category or class. The model learns from these labels to predict categories for new data. This is fundamentally different from clustering, where no labels are provided.

Semi-supervised learning is a hybrid approach that uses a small amount of labeled data combined with a large amount of unlabeled data. While it may sometimes incorporate clustering techniques, clustering itself is purely unsupervised.

Reinforcement learning involves an agent learning to make decisions by receiving rewards or penalties based on actions taken in an environment. This is entirely different from clustering, which focuses on discovering data patterns rather than learning through trial-and-error interactions.

The coefficient of determination (R²) is a statistical metric that measures the proportion of variance in the dependent variable that is predictable from the independent variable(s) in a regression model.

The correct range for R² is between 0 and 1:
- A value of 0 indicates that the model explains none of the variability in the response data around its mean
- A value of 1 indicates that the model explains all the variability in the response data around its mean
- Values closer to 1 represent a better fit, meaning the model's predictions are more accurate
- In practice, an R² of 0.7 or higher is often considered a good fit, though this varies by field and application

Why the other options are incorrect:

The option suggesting a range between -1 and 1 confuses R² with the correlation coefficient (Pearson's r), which does range from -1 to 1 and measures the strength and direction of linear relationships. While R² is related to the correlation coefficient (it's actually the square of r in simple linear regression), R² itself cannot be negative in standard regression scenarios and doesn't indicate inverse relationships.

The option suggesting a range between 0 and 100 incorrectly presents R² as a percentage scale. While R² can be expressed as a percentage (by multiplying by 100), the metric itself is measured on a 0 to 1 scale, not 0 to 100.

The option suggesting a range between 0 and infinity with lower values being better describes metrics like Mean Squared Error (MSE) or Root Mean Squared Error (RMSE), which are loss metrics where lower values indicate better performance. This is the opposite of R², where higher values indicate better fit.