The 5 Regression Metrics That Matter: Last-Minute ML Interview Prep

Author(s): Raghu Teja Manchala

Originally published on Towards AI.

Short and Concise: The Most Asked Regression Metrics in Interviews.

Over the past few years, I have had numerous interviews, ranging from scenario-based to technical rounds. When it comes to machine learning regression models, interviewers typically focus on five key performance metrics, which are the ones mostly used by Data Scientists in real time.

In this article, I have explained each of these key metrics in a short and concise way, using real-life examples to make them easy to understand. This will help you apply these concepts in real-world scenarios and answer interview questions accurately, meeting the interviewer’s expectations.

Introduction

Model performance metrics are a crucial component of the Machine Learning lifecycle that comes after model training.

• Assess model performance.
• Measure how accurately the model predicts on new, unseen data.
• Provide insights into the model’s strengths and weaknesses.
• Help compare different models to choose the best one.

Regression Metrics

1. Mean Squared Error (MSE):

The average of squared differences between predicted and actual values.

• It measures how far the model’s predictions deviate from the actual values.

👉 Useful for penalizing large errors more heavily.
👉 Smaller MSE = Better predictions.

Example: Stock Price prediction

2. Root Mean Squared Error (RMSE):

The square root of the average of squared differences between predicted and actual values.

• It measures how far the model’s predictions deviate from actual values, expressed in the same units as the target variable.

👉 Useful for evaluating model performance in the same units as the target.
👉 Smaller RMSE = Better predictions.
👉 Easy to interpret and explain.

Example: Plant Height prediction
👉 If the RMSE is 3 cm, It means the average difference between predicted and actual plant heights is about 3cm.

3. Mean Absolute Error (MAE):

The average of absolute differences between predicted and actual values.

• It measures how far the model’s predictions deviate from actual values, without considering whether the errors are positive or negative.

👉 Useful for determining the average error in the same units as the target variable.
👉 Less sensitive to outliers.
👉 Smaller MAE = Better predictions.

Example: House Price prediction

4. R-Squared Score (R2):

It is also called the “Coefficient of Determination” which measures the proportion of variance or information in the target variable that can be explained by the model.

• It shows how well the model’s predictions match the actual data.

👉 It evaluates the overall performance of the model, with values ranging from 0 to 1.
👉 Higher R2 = Better predictions.

Example: House Price prediction
While predicting house prices, If the R2 score is 0.85, It means the model explains 85% of variance or information in house prices.

Problem with R2:
👉 It doesn’t consider the correlation between dependent (target) and independent (input) features.
👉 Adding more input features will blindly increase the R2 value, making the model appear to perform better than it actually does.
👉 The regression model tries to assign coefficients in such a way that the sum of squared residuals (ss_res) always decreases.

5. Adjusted R-Squared Score (Adjusted R2):

• It is a modified version of the R2 Score which considers the number of input features used to predict the target variable.
• It helps determine whether adding new input features to the model actually improves its fit.

→ R2: R-Squared Score determined by the model.
→ N: Total number of data points.
→ P: The number of input features.

👉 It penalizes the model for adding features that are not correlated with the target variable.
👉 Higher Adjusted R2 = Better predictions.

Example: House Price prediction

Note: If adding a new feature increases Adjusted R2, It means the feature improves the model otherwise the feature is not adding much value (an unnecessary feature).

Conclusion:

The five regression metrics discussed above are among the most commonly used in real-world applications. Understanding these metrics and selecting the appropriate ones based on the specific business problem and data characteristics is crucial for effectively evaluating regression models.

For a Data Scientist, These metrics are a key part of building models and come up often in daily work. As a result, they are commonly discussed in interviews.

Thank you for reading. I hope this helps with your interview preparation and job role. Feel free to comment with any questions or feedback.

If you like the article and would like to support me, make sure to:

📰 Follow me and explore more content on my medium profile

👏 Give 50 Claps to help this story reach a wider audience.

🔔 Connect with me on LinkedIn

Wishing you a joyful and successful learning journey! 🤝 Let’s Grow Together!

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI