为什么我的逻辑回归只产生一个 class？

Question

I attempted my first machine learning project using a fictional dataset from Kaggle consisting of 1470 records. 84% of records were in the '0' class and 16% were '1's. I used 1200 records to train and test and saved 270 to feed in as new data to see what would happen. I ended up with a training score of 87% and a test score of 83%, but all 270 records of new data were classified as 0.

Could it be that the data, being fictional, just doesn't make enough of a pattern to teach the machine how to classify? Or am I doing something wrong?

I've read some of the other posts that seem to concern a similar problem but I don't find a relevant response. Any help would be appreciated.

df=pd.read_csv('Resources/train_data.csv')
    
df_skinny =df.drop(['EducationField','EmployeeCount','EmployeeNumber','index',
    'StandardHours', 
    'JobRole','MaritalStatus','DailyRate','MonthlyRate','HourlyRate','Over18','OverTime'], 
    axis=1).drop_duplicates()
    df_skinny.rename(columns={"Attrition": "EmploymentStatus"}, inplace=True)
    df_skinny['EmploymentStatus'] = df_skinny['EmploymentStatus'].replace(['Yes','No'],[1,0])

df_skinny['Gender']=df_skinny['Gender'].replace(['Female','Male'],[0,1]) df_skinny['BusinessTravel'] = df_skinny['BusinessTravel'].replace(['Travel_Rarely','Travel_Frequently','Non-Travel'],[1,2,0]) df_skinny['Department']=df_skinny['Department'].replace(['Human Resources','Sales','R&D'],[0,1,2])

df_train=df_skinny[:1200]
df_new=df_skinny[1201:]

X =df_train.drop("EmploymentStatus", axis=1)
y = df_train["EmploymentStatus"]

from sklearn.model_selection import train_test_split

X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=1)

from sklearn.preprocessing import StandardScaler
X_scaler = StandardScaler().fit(X_train)
X_train_scaled = X_scaler.transform(X_train)
X_test_scaled = X_scaler.transform(X_test)

from sklearn.linear_model import LogisticRegression
classifier = LogisticRegression()

classifier.fit(X_train_scaled, y_train)

print(f"Training Data Score: {classifier.score(X_train_scaled, y_train)}")
print(f"Testing Data Score: {classifier.score(X_test_scaled, y_test)}")

predictions = classifier.predict(X_test_scaled)
print(f"First 30 Predictions:   {predictions[:30]}")
print(f"First 30 Actual Employment Status: {y_test[:30].tolist()}")

new_X = df_new.drop("EmploymentStatus", axis=1)
new_predictions=classifier.predict(new_X)
print(new_predictions)

ynew = classifier.predict_proba(new_X)
print(ynew)

OUTPUT:
Training Data Score: 0.8655555555555555
Testing Data Score: 0.8333333333333334

First 30 Predictions:   [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 0 0 0 0 0 0 0 0]

First 30 Actual Employment Status: [1, 1, 1, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 0, 0, 0] 

[0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 0 0 0 0 0 0]

[[1.00000000e+000 0.00000000e+000]
 [1.00000000e+000 0.00000000e+000]
 [1.00000000e+000 0.00000000e+000]
 [1.00000000e+000 0.00000000e+000]
 [1.00000000e+000 5.24119991e-298]
 [1.00000000e+000 7.88999798e-158]
 [1.00000000e+000 2.73485216e-286]
 [1.00000000e+000 0.00000000e+000]
 [1.00000000e+000 0.00000000e+000]

Answer 1

As you have mentioned that 84% of the data is in class 0 and 16% is in class 1. This is very unbalanced data and model will be very biased in this case. That's why you are getting the results mostly as 0.

A good dataset is something that has balanced data among all the classes. So you need to make it balanced using Random sampling techniques. There are two kinds of sampling oversampling and undersampling .

I recommend you to apply sampling techniques to balance your data first.

You can learn more about it from the below article https://machinelearningmastery.com/random-oversampling-and-undersampling-for-imbalanced-classification/

You can refer this notebook https://www.kaggle.com/shweta2407/oversampling-vs-undersampling-techniques