Online STEM and Robotics Coding Club for Can Tho Students

1. Prediction Error Is Inevitable

Linear regression (and ML models in general) never achieve perfect predictions on every real-world case. Error must be measured and shared with clients as an expected margin—similar to a product label that reads 1000g ± 10g.

2. Calculators vs Learning Models

• Calculator: hardcoded rules → exact, error-free answers; no learning
• ML model: discovers rules from historical data and applies them to unseen cases → a residual error margin remains

3. What a Cost Function Is

The cost function is the mathematical metric that quantifies the difference between model predictions and ground-truth values across a dataset. On each test example, prediction vs actual yields an error; the cost function aggregates those errors into one score.

4. Why Simple Sums Fail

Adding raw errors can cancel positives and negatives. Example: errors of +4 and −4 sum to 0, which wrongly suggests a perfect model even though both predictions were wrong.

5. Mean Squared Error (MSE)

1. Compute each prediction error
2. Square every error (negatives become positive)
3. Average the squared values

MSE yields a single non-negative magnitude that reflects overall model error without cancellation.