Questions
3 questions per paper
Difficulty
Medium-Hard
Importance
High yield for core computer science semester exams
Overview
Supervised learning is a paradigm of machine learning where models are trained using labeled datasets, mapping input features to known target outputs. It is a cornerstone of data science exams, focusing on the ability of algorithms to generalize from training data to predict unseen outcomes, making it critical for both descriptive and analytical exam questions.
Regression Analysis
Regression is used for predicting continuous numerical values by establishing a mathematical relationship between independent and dependent variables. It is the fundamental supervised task for quantitative forecasting.
- Linear regression finds the best-fit line: y = mx + c
- Cost function: Mean Squared Error (MSE)
- Goal: Minimize residual sum of squares
- Multivariate regression extends the model to multiple inputs
Classification Techniques
Classification involves categorizing input data into discrete classes, which is essential for pattern recognition and decision-making systems. Exams often test the logic behind splitting criteria and distance metrics.
- Decision Trees: Uses Entropy and Information Gain for node splitting
- k-Nearest Neighbors (kNN): A lazy learner based on distance metrics like Euclidean or Manhattan
- Support Vector Machines (SVM): Maximizes the hyperplane margin between classes
- Kernel Trick: Transforms data into higher dimensions to handle non-linearity
Bias-Variance Tradeoff
This tradeoff describes the balance between a model's complexity and its ability to generalize, which is central to preventing both underfitting and overfitting. It is a high-yield concept for conceptual written exams.
- Bias: Error from erroneous assumptions in the learning algorithm
- Variance: Error from sensitivity to small fluctuations in the training set
- Underfitting: High bias, low variance
- Overfitting: Low bias, high variance
- Goal: Achieve the minimum Total Error point on the U-shaped curve
Formula Sheet
MSE = (1/n) * Σ(yi - ŷi)^2
Entropy = -Σ pi log2(pi)
Euclidean Distance = sqrt(Σ(xi - yi)^2)
Exam Tip
Always draw the Bias-Variance tradeoff U-curve diagram; it acts as a visual anchor that guarantees higher marks in theory-heavy exam papers.
Common Mistakes
- Confusing the purpose of Bias (underfitting) with Variance (overfitting) during explanations.
- Failing to mention the Kernel Trick when asked about SVM limitations.
- Forgetting that kNN is a 'lazy' learner that does not compute a discriminative function during training.
More Revision Notes
Ready to test yourself?
Play topic-wise Supervised Learning questions in Aspirant Arcade — gamified MCQ practice.
Download Free