Deep Learning Series 8 - Classification Model Evaluation Metrics

1 Overview

1.1 Classification

Classification: Labels are discrete values.

Regression: Labels are continuous values.

2. Confusion matrix

Confusion matrix for two categories:

TP and TN are the parts that the network predicted correctly, and FP and FN are the parts that the network predicted incorrectly.

3. Secondary indicators

Accuracy: For the overall evaluation of the model, the proportion of correctly predicted samples to the total number of samples.

Precision rate: Among the predicted results, the proportion of correct predictions;

Recall: The proportion of correct predictions among the real results.

3.1 F1-Score

4. An example

Confusion matrix for three classification problems.

4.1 Accuracy

Converting the problem into a binary classification problem of determining whether it is a cat, the confusion matrix degenerates as follows:

4.2 Accuracy

4.3 Recall rate

4.4 Specificity

4.5 F1-Score

5. MobileNetV2 predicts confusion matrix for flower classification

The abscissa is the true label (True Labels), and the ordinate is the predicted label (Predicted Labels).

Note: In different materials, we can see that the rows and columns of the confusion matrix may be defined differently. Please pay attention to the specific use and analysis.

---

Welcome to follow the public account: [Qianyi Qianxun] and grow together.

---

reference:

1. [Brother Zihao] Basics of Machine Learning
2. Calculate confusion matrix of classification model using pytorch and tensorflow
3. Model effect evaluation—confusion matrix (principle and Python implementation)
4. Deep Learning Image Classification (1) – Confusion Matrix of Classification Model