Back to: Data Science Tutorials

**Model Evaluation for Regression in Machine Learning**

In this article, I am going to discuss **Model Evaluation for Regression in Machine Learning** with Examples. Please read our previous article where we discussed **Model Building and Validation in Machine Learning** with Examples.

**Model Evaluation for Regression in Machine Learning**

We have multiple evaluation metrics for checking the model performance of a regression model –

**Mean Absolute Error –**

The absolute difference between actual and anticipated values is calculated using the MAE measure, which is a relatively basic statistic. Now you must locate your model’s MAE, which is essentially a mistake created by the model and is referred to as an error. Get the difference between the actual and anticipated values, which is an absolute error; however, we must first find the mean absolute of the entire dataset.

Where,

- N = number of data points
- Y = actual output
- Y’ = predicted output

The aim of a regression model should be to get the lowest MAE possible.

**Advantages –**

- The MAE value you receive is in the same unit as the output variable.
- It’s the most resistant to outliers.

**Disadvantages – **Because the graph of MAE is not differentiable, we must use differentiable optimizers such as gradient descent.

**Mean Squared Error –**

MSE is a widely used and straightforward statistic that accounts for a small change in mean absolute error. Finding the squared difference between the actual and anticipated value is defined as a mean squared error.

**What does the MSE actually stand for? **

It denotes the difference in squared values between actual and predicted values. The benefit of MSE is that we conduct squared to avoid the cancellation of negative terms.

Where,

- N = number of data points
- Y = actual output
- Y’ = predicted output

**Advantages – **Because MSE’s graph is differentiable, it can readily be used as a loss function.

**Disadvantages –**

- A squared unit of output is the result of computing MSE. For example, if the output variable is in meter(m), the output we get after computing MSE is in meter squared.
- If the dataset contains outliers, the outliers are penalized the most, and the estimated MSE is larger. In other words, it is not robust against outliers, which was a benefit in MAE.

**Root Mean Squared Error –**

The acronym RMSE indicates that it is a simple square root of mean squared error.

**Advantages –** Because the output value is in the same unit as the desired output variable, loss interpretation is simple.

**Disadvantages –** When compared to MAE, it is less resistant to outliers.

**R-Squared (R2) Error –**

The R2 score is a metric that measures the performance of your model, not the loss in terms of how many wells it performed. As we’ve seen, MAE and MSE are context-dependent, whereas the R2 score is context-independent.

So, using R squared, we can compare a model to a baseline model that none of the other metrics can provide. In classification problems, we have something similar called a threshold, which is set at 0.5. R2 squared calculates how much better a regression line is than a mean line.

Where,

- SSr = Squared sum error of regression line
- SSm = Squared sum error of mean line

The R2 value increases as our regression line approach perfection. Furthermore, the model’s performance improves.

In the next article, I am going to discuss **Classification and its Use Cases in Machine Learning** with Examples. Here, in this article, I try to explain **Model Evaluation for Regression in Machine Learning** with Examples. I hope you enjoy this Model Evaluation for Regression in Machine Learning with Examples article.