Linear regression concept in ML Python - Model Metrics & Evaluation

Metrics & Evaluation - Linear regression concept

Which metric matters for Linear Regression and WHY

For linear regression, we want to measure how close our predicted numbers are to the real numbers. The most common metrics are Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE). These metrics tell us the average size of the mistakes our model makes.

MSE and RMSE give more weight to big mistakes because they square the errors. MAE treats all mistakes equally. We choose these because linear regression predicts continuous values, so accuracy or classification metrics don't apply.

Confusion Matrix or Equivalent Visualization

Linear regression does not use a confusion matrix because it predicts numbers, not categories. Instead, we look at error values.

Example errors for 5 predictions:
Actual:    [3, 5, 7, 9, 11]
Predicted: [2.5, 5.5, 6.8, 9.2, 10.5]
Errors:    [-0.5, 0.5, -0.2, 0.2, -0.5]

MSE = (0.5² + 0.5² + 0.2² + 0.2² + 0.5²) / 5 = 0.17
RMSE = sqrt(0.17) ≈ 0.41
MAE = (0.5 + 0.5 + 0.2 + 0.2 + 0.5) / 5 = 0.38
    

Tradeoff: Choosing the Right Error Metric

If you want to avoid big mistakes, use MSE or RMSE because they punish large errors more. For example, if predicting house prices, a big mistake can cost a lot.

If you want a simple average error size, use MAE. It is easier to understand and less sensitive to outliers.

Choosing the right metric depends on what mistakes matter more in your problem.

What Good vs Bad Metric Values Look Like

Good values mean small errors. For example, if house prices are around $200,000, an RMSE of $1,000 is very good. It means predictions are close.

Bad values mean large errors. An RMSE of $50,000 means predictions are often far off, which is not useful.

Always compare error size to the range of your data to understand if it is good or bad.

Common Pitfalls in Linear Regression Metrics

• Ignoring scale: Reporting error without considering data scale can mislead. A 10 error might be big or small depending on the data.
• Overfitting: Very low training error but high test error means the model learned noise, not patterns.
• Data leakage: Using future or test data in training can give unrealistically low errors.
• Outliers: Can inflate MSE and RMSE, making the model look worse than it is.

Self Check: Is a Model with Low MSE Always Good?

Imagine your model has a very low MSE on training data but much higher error on new data. This means it is overfitting and will not perform well in real life.

Also, if your data values are very large, a low MSE might still mean big mistakes in real terms. Always check errors relative to data scale and test data.

Key Result

Linear regression uses error metrics like MSE, RMSE, and MAE to measure how close predictions are to actual values, focusing on average prediction mistakes.

Practice

1. What does linear regression primarily do in machine learning?

easy

A. It groups data points into clusters.

B. It reduces the number of features in data.

C. It finds a straight line to predict values based on input data.

D. It creates decision trees for classification.

Solution

1. Step 1: Understand the goal of linear regression

   Linear regression aims to find a straight line that best fits the data points to predict continuous values.

2. Step 2: Compare with other methods

   Clustering groups data, feature reduction simplifies data, and decision trees classify data, which are different tasks.

3. Final Answer:

   It finds a straight line to predict values based on input data. -> Option C

4. Quick Check:

   Linear regression = straight line prediction [OK]

Hint: Linear regression fits a line to predict numbers [OK]

Common Mistakes:

• Confusing regression with classification
• Thinking it groups data instead of predicting
• Mixing it up with feature reduction techniques

2. Which of the following is the correct formula for a simple linear regression model?

easy

A. y = m - x / b

B. y = mx + b

C. y = m + x * b

D. y = x / m + b

Solution

1. Step 1: Recall the linear regression formula

   The formula is y = mx + b, where m is slope and b is intercept.

2. Step 2: Check each option

   Only y = mx + b matches the correct formula exactly.

3. Final Answer:

   y = mx + b -> Option B

4. Quick Check:

   Linear formula = y = mx + b [OK]

Hint: Slope times x plus intercept equals y [OK]

Common Mistakes:

• Mixing multiplication and division signs
• Swapping slope and intercept positions
• Using incorrect operators between terms

3. Given the linear regression model y = 2x + 3, what is the predicted value when x = 4?

medium

A. 11

B. 14

C. 8

D. 10

Solution

1. Step 1: Substitute x = 4 into the model

   Calculate y = 2 * 4 + 3.

2. Step 2: Perform the calculation

   2 * 4 = 8, then 8 + 3 = 11.

3. Final Answer:

   11 -> Option A

4. Quick Check:

   2*4+3 = 11 [OK]

Hint: Multiply slope by x, then add intercept [OK]

Common Mistakes:

• Adding before multiplying
• Using wrong slope or intercept values
• Confusing x with y

4. Identify the error in this Python code for linear regression prediction:

def predict(x):
    return 2 * x +

print(predict(5))

medium

A. Syntax error due to incomplete return statement

B. Missing intercept value in the formula

C. Wrong function name for prediction

D. Incorrect multiplication operator

Solution

1. Step 1: Check the return statement syntax

   The return line ends with a plus sign and no value after it, causing a syntax error.

2. Step 2: Confirm error type

   Python requires complete expressions; incomplete return causes syntax error.

3. Final Answer:

   Syntax error due to incomplete return statement -> Option A

4. Quick Check:

   Incomplete return line = syntax error [OK]

Hint: Check if return line ends abruptly [OK]

Common Mistakes:

• Ignoring incomplete expression errors
• Assuming missing intercept is the error
• Thinking function name matters here

5. You have data points: (1,3), (2,5), (3,7). Which linear regression model best fits this data?

hard

A. y = x + 2

B. y = 3x + 0

C. y = 2x + 3

D. y = 2x + 1

Solution

1. Step 1: Calculate slope (m)

   Change in y: 5-3=2, 7-5=2; change in x: 2-1=1, 3-2=1; slope m = 2/1 = 2.

2. Step 2: Calculate intercept (b)

   Use point (1,3): 3 = 2*1 + b -> b = 3 - 2 = 1.

3. Final Answer:

   y = 2x + 1 -> Option D

4. Quick Check:

   Slope=2, intercept=1 fits points [OK]

Hint: Slope = change in y/change in x, intercept from point [OK]

Common Mistakes:

• Mixing slope and intercept values
• Using wrong points for calculation
• Assuming intercept is zero without checking