我如何使用机器学习预测事件发生的机会？

Question

I have a dataset of events including their coordinates, Hour of the day, day of the week, and the weather (temperature and downfall) on that specific day. My goal is to predict the chance of that event to occur when you input these values. So I have a lot of data (about 1500 entries) of occurrences, but obviously, none where it did not happen.

I looked at XGBoost because someone suggested it but I can't really find out how I could use that, how I applied it now just always returns 1.

This is my current implementation

import xgboost as xgb

#RD is my dataframe 

#I labeled everything with a 1 since xgboost needs to predict something. I have no clue how i could handle this better :)
rd["Label"] = 1

X,y=rd[["HourOfDay",'Type','Lat','Long','DayOfWeek']],rd["Label"]
xg_cl = xgb.XGBClassifier(objective="binary:logistic",
                         n_estimators=10, seed=123)
xg_cl.fit(X,y)

testdf = pd.DataFrame({
    "HourOfDay" : [1],
    "Type" : [2],
    "Lat" : [0],
    "Long" : [0],
    "DayOfWeek" : [6]
})
preds=xg_cl.predict(data =testdf)

This code always gives me true (1) but I need it to return the chance of the event happening and I am pretty sure my current implementation is useless.

Can somebody point me in the right direction on how to solve this issue?

Answer 1

When you label every data point with a 1 and train your model on those, of course the classifier will only predict 1 . It has no idea another class even exists. You need different labelled examples in your training data in order to properly fit a classifier.

Besided that, there is a predict_proba function that returns the class probabilities rather than the classes.