Author: Zen and the Art of Computer Programming
Model Evaluation in Machine Learning: From Accuracy to Recall
- introduction
1.1. Background Introduction
Machine learning models are widely used in today's data-driven society, but how to evaluate the performance of models is still a difficult problem for many practitioners. With the development of the Internet and big data technology, the demand for model performance evaluation is becoming stronger and stronger. Traditional indicators such as accuracy and recall rate can no longer meet more and more diverse application scenarios. Therefore, this article will introduce a general model evaluation method from accuracy to recall to help practitioners better evaluate the performance of the model.
1.2. Purpose of the article
This article aims to describe a general model evaluation method from accuracy to recall rate, to help readers understand how to extract useful information from the performance data of the model, so as to provide strong support for the improvement of the model.
1.3. Target audience
This article is suitable for practitioners with certain machine learning foundation and application experience, as well as those who have a certain need for model performance evaluation.
- Technical Principles and Concepts
2.1. Explanation of basic concepts
Recall (Recall): refers to the proportion of correctly identified samples to the total number of samples that are actually of a specific category.
Accuracy: Refers to the proportion of the correct samples predicted by the model to the total number of samples.
2.2. Technical principle introduction: algorithm principle, operation steps, mathematical formulas, etc.
For a given classification problem, we can establish the following mathematical model according to the actual business scenario:
y = f(x)
where y is the sample feature, f( x) is the label predicted by the model, and x is the sample feature.
Then, we collect a certain amount of data sets and preprocess the data sets, such as missing value processing, outlier processing, etc.
Next, train a machine learning model based on the dataset, such as the line