Simple understanding of Precision, Recall, Accuracy, TP, TN, FP, FN

True Positive(TP),True Negative(TN),False Positive(FP),False Negative(FN)
these four words,The left half representsIs the prediction of the model correct?(分为True, False),The right half represents模型预测的结果(分为Positive, Negative).因此有：

• TP：The model predicted right$(T)$,positive indeed$(P)$,即所谓的"真阳性"
• TN：The model predicted right$(T)$,indeed negative$(N)$,即所谓的"真阴性"
• FP：模型预测错了$(F)$,predicted positive$(P)$,So it's actually negative,即所谓的"假阳性"
• FN：模型预测错了$(F)$,预测为阴性$(N)$,So actually positive,即所谓的"假阴性"

---

精确率： All positive predictions given by the model$(P)$中,How much is actually true$(TP)$,即：$$Precision=\frac{TP}{TP+FP}$$ Precision is more focused on measuring whether the model's positive predictions are reliable(Can't kill by mistake).容易发现,假阳性FP越高,less accurate.

---

召回率： of all actual positive samples(TP+FN),How many were correctly discovered by the model(TP),即：$$Recall=\frac{TP}{TP+FN}$$ Recall is more focused on measuring whether actual positive samples are missed(can't miss).容易发现,假阴性FN越高,The lower the recall rate.

---

准确率： Model to predict the correct samples(T)占总样本的比例,即：$$Accuracy=\frac{TP+TN}{TP+TN+FP+FN}$$ Accuracy are often used to measure the comprehensive performance of a machine learning model.不过,Since the proportion of yin and yang samples in practical applications is likely to be unbalanced,Positive samples are generally rare,The model can achieve a high accuracy rate by simply predicting all samples as negative.为了解决这一问题,precision rates that focus more on the positivity itself,The two metrics of recall.