Untitled

import pandas as pd
import numpy as np
from sklearn.model_selection import StratifiedKFold
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score, roc_auc_score, average_precision_score
from sklearn.ensemble import RandomForestClassifier
from sklearn.linear_model import LogisticRegression
import xgboost as xgb
import lightgbm as lgb

# Copy the dataset to avoid modifying the original
df = final_train_data.copy()

# Binning continuous variables for stratification
df['TransactionAmt_bin'] = pd.qcut(df['TransactionAmt'], q=5, labels=False, duplicates='drop')
df['TransactionDT_bin'] = pd.qcut(df['TransactionDT'], q=5, labels=False, duplicates='drop')

# Convert TransactionDT into useful time-based features
df["TransactionDT_days"] = df["TransactionDT"] // (24 * 60 * 60)  # Convert to days
df["TransactionDT_hours"] = (df["TransactionDT"] // 3600) % 24  # Extract hour of day
# Categorical columns to balance
categorical_cols = ['ProductCD', 'card1', 'card2', 'card3', 'card4', 'card5',
                    'DeviceType', 'DeviceInfo', 'P_emaildomain', 'R_emaildomain']

# Reduce category explosion by grouping rare values
for col in categorical_cols:
    df[col] = df[col].astype(str)  # Ensure strings
    top_categories = df[col].value_counts().index[:10]  # Keep only the top 10 most frequent
    df[col] = df[col].apply(lambda x: x if x in top_categories else "Other")  # Group rare categories

# Adjust rare fraud cases by ensuring at least `n_splits` samples exist per category
df['fraud_count'] = df.groupby('isFraud')['isFraud'].transform('count')
df.loc[df['fraud_count'] < 5, 'isFraud'] = df['isFraud'].replace({1: 0})  # Reassign rare fraud cases

# Create a stratification label combining key features
df['stratify_label'] = (
    df['isFraud'].astype(str) + "_" +  # Fraud balance
    df['TransactionAmt_bin'].astype(str) + "_" +  # Transaction amount range
    df['TransactionDT_bin'].astype(str) + "_" +  # Transaction time range
    df['ProductCD'] + "_" +  # Product type
    df['card1'] + "_" +  # Card type
    df['card2'] + "_" +
    df['card3'] + "_" +
    df['card4'] + "_" +
    df['card5'] + "_" +
    df['card6'] + "_" +
    df['DeviceType']  # Device variety
)