The only Performance Metrics article you will ever need!

Performance metrics are a crucial part of machine learning, as they allow us to evaluate the quality of a model and make informed decisions about how to improve it. In this article, we will explore most of the commonly used performance metrics and the math behind them. In addition, it also provides a simple Numpy implementation of all the metrics which can be tailored to suit your needs.

Error Metrics

If expressed in simple layman terms, an error in machine learning is basically the deviation of the predicted value from the original value. We seek to measure the error terms for all the records present in our data using various techniques such as:

Mean Error

Mean error (ME) is the mean of the errors.

• It is used to evaluate the overall accuracy of a model’s predictions.
• Mean error is sensitive to outliers, as it is calculated by summing the errors and dividing by the total number of predictions. A few extreme errors can greatly influence the result.
• Mean error is useful for understanding the average magnitude and direction of the errors, but it can be misleading if there are large outliers in the data.

It is calculated as follows:

Mean Percentage Error

Mean percentage error (MPE) is the percentage of the sum of the ratio of predicted errors to actual target…