Home/Notes/Model Evaluation & Optimization
Board Exam Notes

Model Evaluation & Optimization Notes

Questions

2–3 questions in university theory papers

Difficulty

Medium

Importance

Core topic for Data Science and AI modules

Overview

Model Evaluation and Optimization are fundamental pillars of machine learning that quantify predictive performance and refine model convergence. Understanding these concepts is vital for university exams as they bridge the gap between theoretical algorithm design and practical system reliability.

Confusion Matrix

A confusion matrix is a tabular representation used to describe the performance of a classification model by comparing actual vs predicted labels. It serves as the foundation for calculating all critical classification performance metrics.

  • True Positives (TP): Correct positive predictions
  • True Negatives (TN): Correct negative predictions
  • False Positives (FP): Type I error
  • False Negatives (FN): Type II error
  • Matrix size is N x N for N classes

Precision and Recall

These metrics provide a nuanced view of performance beyond simple accuracy, especially in imbalanced datasets. Precision focuses on exactness, while Recall measures the ability of the model to capture all relevant instances.

  • Precision = TP / (TP + FP)
  • Recall = TP / (TP + FN)
  • F1-Score = 2 * (Precision * Recall) / (Precision + Recall)
  • High precision implies low false positive rate
  • High recall implies low false negative rate

Cross-Validation

Cross-validation is a statistical technique used to estimate the skill of machine learning models on unseen data by partitioning the data into subsets. It effectively mitigates the risk of overfitting by ensuring the model performs consistently across different data segments.

  • K-Fold Cross-Validation: Data split into K equal folds
  • Leave-One-Out Cross-Validation (LOOCV)
  • Stratified K-Fold maintains class distributions
  • Reduces variance in performance estimation
  • Uses training data for validation loop

Gradient Descent

Gradient descent is an iterative optimization algorithm used to minimize the cost function by adjusting model parameters in the direction of the steepest descent. It is the primary method for training neural networks and linear regression models.

  • Learning Rate (alpha) controls step size
  • Batch Gradient Descent uses full dataset
  • Stochastic Gradient Descent (SGD) uses single samples
  • Convergence occurs when cost reaches local minimum
  • Weight update: W_new = W_old - alpha * gradient

Formula Sheet

Accuracy = (TP + TN) / (TP + TN + FP + FN)

F1-Score = 2PR / (P + R)

Gradient Descent Update: θ = θ - α * ∇J(θ)

Exam Tip

Always draw the 2x2 Confusion Matrix grid in your answer, even if the question only asks for the definitions of Precision and Recall; examiners award significant marks for visual representation.

Common Mistakes

  • Confusing Type I (FP) and Type II (FN) errors in definitions
  • Selecting a learning rate that is too high, leading to divergence instead of convergence
  • Failing to mention the trade-off between Precision and Recall in short-answer questions

More Revision Notes

Ready to test yourself?

Play topic-wise Model Evaluation & Optimization questions in Aspirant Arcade — gamified MCQ practice.

Download Free