Confusion matrix learning notes

1. Conceptual understanding ：
   1. The confusion matrix is ROC The basis of curve drawing , It is also the most basic way to measure the accuracy of the classification model , Most intuitive , The simplest way to calculate
   2. Simple understanding ： The confusion matrix is the error classification of the statistical classification model , The number of observations in the corresponding category is displayed in a table
   3. Confusion matrix is often used to judge the advantages and disadvantages of classifiers .
   4. extend ： Common methods of judging by type models ：
      1. Confusion matrix
      2. ROC curve
      3. AUC area
2. For the first level indicators of the second classification ：(positive Short for p,negative Short for n）
   1. TP： The real value is p, The prediction is p The number of
   2. FN( The second kind of statistical error )： The real value is p, The prediction is n The number of
   3. FP( The first kind of statistical error )： The real value is n, The prediction is p The number of
   4. TN： The real value is n, The prediction is n The number of
   5. Ingenious notes ： It's all about judging the predicted value , If the prediction is correct, it will be T, If the prediction is wrong F
   6. When you get the matrix, you will hope TP and TN Bigger , That is, the value on the sub diagonal is greater
3. Secondary indicators of the second classification ：（ Put forward the reason ： The statistics in the confusion matrix are numbers , For a lot of data , It is impossible to measure the merits of the model according to the number ）

    

   1. Accuracy rate ： Calculate the proportion of all correctly judged results in the total predicted value
   2. accuracy ： Forecast as p The correct proportion
   3. sensitivity ： Recall rate , The true value is p Predict the correct proportion in the results
   4. Specificity ： The real value is n The proportion of the predicted pair in the model
   5. The formula ：

   6. adopt 4 Two indicators , You can convert the value to 0-1 The ratio between , Facilitate standardized measurement

4. Third level index ：F1 score= Accuracy （precision） Reciprocal + Recall rate （recall） Reciprocal , Value range [0,1],1 The output representing the model is the best ,0 The output of representative model is the worst .
5. For multi classification , It can be decomposed into multiple binary classifications （ yes or No ）, Or it can be directly integrated into a large matrix