Credit-Card-Anomaly / utils /preprocessing.py
Zayeemk's picture
Rename preprocessing.py to utils/preprocessing.py
9a8c220 verified
Raw History Blame Contribute Delete
12.3 kB
import pandas as pd
import numpy as np
from sklearn.preprocessing import StandardScaler, LabelEncoder
from typing import Tuple, Dict, Optional
import os
import logging
# Configure logging
logging.basicConfig(level=logging.ERROR)
logger = logging.getLogger(__name__)
class DataPreprocessor:
"""Handles data preprocessing for credit card anomaly detection."""
def __init__(self):
self.scaler = StandardScaler()
self.label_encoders = {}
self.user_stats = {}
def load_data(self, filepath: str) -> pd.DataFrame:
"""Load data from multiple file formats (CSV, Excel, JSON, Parquet)."""
try:
file_ext = os.path.splitext(filepath)[1].lower()
if file_ext == '.csv':
df = pd.read_csv(filepath)
elif file_ext in ['.xlsx', '.xls']:
df = pd.read_excel(filepath)
elif file_ext == '.json':
df = pd.read_json(filepath)
elif file_ext == '.parquet':
df = pd.read_parquet(filepath)
else:
# Try CSV as fallback
try:
df = pd.read_csv(filepath)
except:
raise ValueError(f"Unsupported file format: {file_ext}")
# Ensure dataframe is not empty
if df.empty:
raise ValueError("Loaded data is empty")
return df
except Exception as e:
logger.error(f"Error loading data: {str(e)}")
# Return empty dataframe with expected columns as fallback
return self._create_empty_dataframe()
def _create_empty_dataframe(self) -> pd.DataFrame:
"""Create an empty dataframe with expected columns as fallback."""
columns = ['Transaction ID', 'User ID', 'Amount', 'Timestamp', 'Merchant Category', 'Location']
return pd.DataFrame(columns=columns)
def ensure_required_columns(self, df: pd.DataFrame) -> pd.DataFrame:
"""Ensure all required columns exist, auto-generate missing ones."""
df = df.copy()
# Define required columns and their default generators
required_columns = {
'Transaction ID': lambda: [f'TX{i:06d}' for i in range(len(df))],
'User ID': lambda: [f'USER{i%10+1:03d}' for i in range(len(df))],
'Amount': lambda: [0.0] * len(df),
'Timestamp': lambda: pd.date_range(start='2024-01-01', periods=len(df), freq='H'),
'Merchant Category': lambda: ['Unknown'] * len(df),
'Location': lambda: ['Unknown'] * len(df)
}
# Add missing columns with generated defaults
for col, generator in required_columns.items():
if col not in df.columns:
try:
df[col] = generator()
except Exception as e:
logger.error(f"Error generating column {col}: {str(e)}")
df[col] = 'Unknown' if col in ['Merchant Category', 'Location'] else 0
# Ensure numeric columns are numeric
numeric_cols = ['Amount']
for col in numeric_cols:
if col in df.columns:
try:
df[col] = pd.to_numeric(df[col], errors='coerce').fillna(0)
except:
df[col] = 0
# Ensure timestamp is datetime
if 'Timestamp' in df.columns:
try:
df['Timestamp'] = pd.to_datetime(df['Timestamp'], errors='coerce')
df['Timestamp'] = df['Timestamp'].fillna(pd.Timestamp('2024-01-01'))
except:
df['Timestamp'] = pd.Timestamp('2024-01-01')
return df
def handle_missing_values(self, df: pd.DataFrame) -> pd.DataFrame:
"""Handle missing values in the dataset with safe operations."""
df = df.copy()
try:
# Fill missing numeric values with median
numeric_cols = df.select_dtypes(include=[np.number]).columns
for col in numeric_cols:
median_val = df[col].median()
if pd.isna(median_val):
median_val = 0
df[col] = df[col].fillna(median_val)
# Fill missing categorical values with mode
categorical_cols = df.select_dtypes(include=['object', 'category']).columns
for col in categorical_cols:
mode_val = df[col].mode()
if not mode_val.empty:
df[col] = df[col].fillna(mode_val[0])
else:
df[col] = df[col].fillna('Unknown')
except Exception as e:
logger.error(f"Error handling missing values: {str(e)}")
# Fallback: fill all NaN with appropriate defaults
for col in df.columns:
if df[col].dtype in [np.number, 'int64', 'float64']:
df[col] = df[col].fillna(0)
else:
df[col] = df[col].fillna('Unknown')
return df
def normalize_per_user(self, df: pd.DataFrame) -> pd.DataFrame:
"""Normalize spending amounts per user with error handling."""
df = df.copy()
try:
# Check if User ID column exists
if 'User ID' not in df.columns:
df['User ID'] = 'USER001'
logger.warning("User ID column missing, using default")
# Check if Amount column exists
if 'Amount' not in df.columns:
df['Amount'] = 0
logger.warning("Amount column missing, using default")
# Calculate user statistics
self.user_stats = df.groupby('User ID')['Amount'].agg(['mean', 'std', 'median']).to_dict('index')
# Normalize amount relative to user average
def normalize_row(row):
user_id = row.get('User ID', 'USER001')
amount = row.get('Amount', 0)
if user_id in self.user_stats:
mean = self.user_stats[user_id]['mean']
std = self.user_stats[user_id]['std']
return (amount - mean) / (std + 1e-8) if std > 0 else 0
else:
return 0
df['Amount_Normalized'] = df.apply(normalize_row, axis=1)
except Exception as e:
logger.error(f"Error normalizing per user: {str(e)}")
# Fallback: use simple z-score
try:
mean = df['Amount'].mean()
std = df['Amount'].std()
df['Amount_Normalized'] = (df['Amount'] - mean) / (std + 1e-8) if std > 0 else 0
except:
df['Amount_Normalized'] = 0
return df
def encode_categorical(self, df: pd.DataFrame) -> pd.DataFrame:
"""Encode categorical variables with error handling."""
df = df.copy()
try:
# Try to find merchant category column with various possible names
category_col = None
possible_names = ['Merchant Category', 'merchant_category', 'Merchant_Category', 'category', 'Category']
for name in possible_names:
if name in df.columns:
category_col = name
break
# If no category column found, use a default
if category_col is None:
if 'Merchant Category' not in df.columns:
df['Merchant Category'] = 'Unknown'
category_col = 'Merchant Category'
# Encode the category column
if category_col in df.columns:
if category_col not in self.label_encoders:
self.label_encoders[category_col] = LabelEncoder()
df[category_col + '_Encoded'] = self.label_encoders[category_col].fit_transform(df[category_col].astype(str))
else:
df[category_col + '_Encoded'] = self.label_encoders[category_col].transform(df[category_col].astype(str))
except Exception as e:
logger.error(f"Error encoding categorical variables: {str(e)}")
# Fallback: add a simple encoded column
if 'Merchant Category_Encoded' not in df.columns:
df['Merchant Category_Encoded'] = 0
return df
def scale_features(self, df: pd.DataFrame, feature_cols: list) -> Tuple[pd.DataFrame, np.ndarray]:
"""Scale numerical features using StandardScaler with error handling."""
df = df.copy()
try:
# Filter to only columns that exist
valid_cols = [col for col in feature_cols if col in df.columns]
if not valid_cols:
logger.warning("No valid columns to scale")
return df, np.array([])
if len(df) > 0:
# Ensure all columns are numeric
for col in valid_cols:
if df[col].dtype not in [np.number, 'int64', 'float64']:
try:
df[col] = pd.to_numeric(df[col], errors='coerce').fillna(0)
except:
df[col] = 0
scaled_features = self.scaler.fit_transform(df[valid_cols])
for i, col in enumerate(valid_cols):
df[col + '_Scaled'] = scaled_features[:, i]
else:
scaled_features = np.array([])
except Exception as e:
logger.error(f"Error scaling features: {str(e)}")
# Fallback: return original dataframe with empty scaled features
scaled_features = np.array([])
return df, scaled_features
def preprocess_pipeline(self, df: pd.DataFrame) -> Tuple[pd.DataFrame, np.ndarray, list]:
"""Complete preprocessing pipeline with comprehensive error handling."""
try:
# Ensure required columns exist
df = self.ensure_required_columns(df)
# Handle missing values
df = self.handle_missing_values(df)
# Normalize per user
df = self.normalize_per_user(df)
# Encode categorical variables
df = self.encode_categorical(df)
# Select feature columns for scaling - only include columns that exist
feature_cols = []
if 'Amount_Normalized' in df.columns:
feature_cols.append('Amount_Normalized')
if 'Merchant_Category_Encoded' in df.columns:
feature_cols.append('Merchant_Category_Encoded')
# Add additional features if available
if 'Amount' in df.columns:
feature_cols.append('Amount')
# If no features selected, use Amount as fallback
if not feature_cols and 'Amount' in df.columns:
feature_cols = ['Amount']
# Scale features only if we have valid columns
if feature_cols:
df, scaled_features = self.scale_features(df, feature_cols)
else:
scaled_features = np.array([])
return df, scaled_features, feature_cols
except Exception as e:
logger.error(f"Error in preprocessing pipeline: {str(e)}")
# Fallback: return original dataframe with minimal processing
try:
df = self.ensure_required_columns(df)
feature_cols = ['Amount']
df, scaled_features = self.scale_features(df, feature_cols)
return df, scaled_features, feature_cols
except:
# Ultimate fallback
return df, np.array([]), ['Amount']
def get_user_statistics(self, user_id: str) -> Dict:
"""Get statistics for a specific user."""
return self.user_stats.get(user_id, {})