如何修复此分类报告警告？

Question

I created a model for multiclass classification. Everything went good, got a validation accuracy of 84% but when I printed the classification report I got this warning:

 UndefinedMetricWarning: Precision and F-score are ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.
  _warn_prf(average, modifier, msg_start, len(result))

classification report:

              precision    recall  f1-score   support

           0       0.84      1.00      0.91     51890
           1       0.67      0.04      0.08      8706
           2       0.00      0.00      0.00      1605

    accuracy                           0.84     62201
   macro avg       0.50      0.35      0.33     62201
weighted avg       0.79      0.84      0.77     62201

Source Code -

import pandas as pd

df=pd.read_csv('Crop_Agriculture_Data_2.csv')
df=df.drop('ID',axis=1)

dummies=pd.get_dummies(df[['Crop_Type', 'Soil_Type', 'Pesticide_Use_Category', 'Season']],drop_first=True)
df=df.drop(['Crop_Type', 'Soil_Type', 'Pesticide_Use_Category', 'Season'],axis=1)
df=pd.concat([df,dummies],axis=1)

df['Crop_Damage']=df['Crop_Damage'].map({'Minimal Damage':0,'Partial Damage':1,'Significant Damage':2})

x=df.drop('Crop_Damage',axis=1).values
y=df.Crop_Damage.values
from sklearn.model_selection import train_test_split
x_train,x_test,y_train,y_test=train_test_split(x,y,train_size=0.3,random_state=101)

from sklearn.preprocessing import MinMaxScaler
mms=MinMaxScaler()
x_train=mms.fit_transform(x_train)
x_test=mms.transform(x_test)

from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense,Dropout,Flatten

model=Sequential()
model.add(Flatten())
model.add(Dense(10,activation='relu'))
model.add(Dropout(0.3))
model.add(Dense(6,activation='relu'))
model.add(Dropout(0.3))
model.add(Dense(3,activation='softmax'))

model.compile(loss='sparse_categorical_crossentropy',optimizer='adam',metrics=['accuracy'])
model.fit(x_train,y_train,validation_data=(x_test,y_test),epochs=13)

import numpy as np
pred=np.argmax(model.predict(x_test),axis=-1)

from sklearn.metrics import classification_report
print(classification_report(y_test,pred))

I think it might be because most of the data is in one category but I'm not sure. Is there anything I can do to solve this?

Answer 1

You don't want to get rid this warning as it says that your class 2 are not on the predictions as there were no samples in the training set

you got an imbalance classification problem and the class 2 has realy low number of samples, and it was present in the test data only

I suggest you 2 things

StratifiedKFold So when you split for training and test, it consider all classes

Oversampling you might need adjust your data by randomly resample the training dataset to duplicate examples from the minority class

Answer 2

As desternaut said, you have a warning , not an error.

This warning is saying you that the classification_report output is influenced because of one of labels is never predicted for your model (in your case, label "2").

This will generate a problem calculating Precision (dividing by 0) , because ( true positives + false positives =0 ). When the function deals with this problem, aoutomatically output 0. Note this is not the real value, it should be "undefined" or something like this, but it's his approach. As you can see, when you are calculating macro avg, you are using this calculated 0. So the error is just reminding you that you macro avg is influenced by a "fake" 0.

The same happens with F1-score, since it is calculated starting from Precision.

How to solve that? Well, technically you don't have nothin to solved because it's not an eror, so you can deal with it. But you have to be aware that your output is being influenced.

What you can do, is decide that you are not interested in the scores of labels that were not predicted, and then explicitly specify the labels you are interested in (which are labels that were predicted at least once):

print(classification_report(y_test,pred,labels=np.unique(y_pred))

Note that this solution is not good at all because it's hidden problems you have with your model and data, but it can be useful in some cases.

Moreover, as Yefet said, your model seems to have problem classifying label "2" because you have an unbalanced data. Follow his suggestions and improve your model if you can.

Answer 3

If you only want to get rid of the warning even knowing that it's hiding problems, you could use zero_division parameter.

According to the documentation :

zero_division: “warn”, 0 or 1, default=”warn”

Sets the value to return when there is a zero division. If set to “warn”, this acts as 0, but warnings are also raised.

So you could hide warnings without altering the result of the classification report with:

print(classification_report(y_test,pred, zero_division=0))

Answer 4

I had the same problem and the solution presented above worked:

classification_report(y_test, y_pred, labels=np.unique(y_pred))

But after thoroughly checking the data, I came to the conclusion that some columns in my data where too large and needed to be scaled/normalised.

Therefore, adding the following scaling to the code provides much better results (at least in my case).

from sklearn.preprocessing import StandardScaler
sc = StandardScaler()
X_train = sc.fit_transform(X_train)
X_test = sc.transform(X_test)