i am trying to make a fraud detection model using an imbalanced data set with fraud being so much higher than un fraud the data set is as follow

RangeIndex: 628343 entries, 0 to 628342
Data columns (total 8 columns):
 #   Column       Non-Null Count   Dtype  
---  ------       --------------   -----  
 0   Time         628343 non-null  object 
 1   Card Number  628343 non-null  float64
 2   merchant     628343 non-null  object 
 3   category     628343 non-null  object 
 4   Amount       628343 non-null  float64
 5   firstName    628343 non-null  object 
 6   lastName     628343 non-null  object 
 7   is_fraud     628342 non-null  float64

i tried to preprocess the data first by dropping the merchant,name , category,trans_num.card num fields and to convert time from this format 1/1/2019 0:00 to hour,second,year,month,day and dropping time then using logistic regression but the accuracy is low any ideas how to preprocess the data and what algorithm to use?