fraud-autoencoder-pyod

Lightweight Python project for fraud/anomaly detection using an AutoEncoder from PyOD.

Overview

This repository provides code and utilities to train, evaluate, and serve an autoencoder-based anomaly detector for tabular transaction data using PyOD.

Features

• Train AutoEncoder (PyOD) on transaction data
• Evaluation scripts: ROC AUC, precision, recall, F1
• Simple preprocessing pipeline (scaling, categorical encoding)
• Save/load trained models
• CLI-style example scripts for train / eval / predict

Repository layout

• data/ — (optional) raw and processed datasets
• notebooks/ — exploratory analysis and examples
• src/ — training, evaluation, and inference code
• models/ — saved model artifacts
• requirements.txt
• README.md

Requirements

• Python 3.8+
• pip

Install dependencies:

pip install -r requirements.txt
# or minimal
pip install numpy pandas scikit-learn pyod joblib

Quickstart example (Python)

Basic training + evaluation flow:

import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from pyod.models.auto_encoder import AutoEncoder
from sklearn.metrics import roc_auc_score, precision_recall_fscore_support
import joblib

# load data: X (features), y (binary labels: 1=anomaly/fraud, 0=normal)
df = pd.read_csv("data/transactions.csv")
X = df.drop(columns=["label"])
y = df["label"].values

# split & preprocess
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42, stratify=y)
scaler = StandardScaler().fit(X_train)
X_train_s = scaler.transform(X_train)
X_test_s = scaler.transform(X_test)

# train
model = AutoEncoder(hidden_neurons=[64, 32, 32, 64], epochs=50, batch_size=64, contamination=0.01)
model.fit(X_train_s)

# predict & evaluate
scores = model.decision_function(X_test_s)  # higher => more abnormal
y_pred = model.predict(X_test_s)            # 1 for outlier

print("ROC AUC:", roc_auc_score(y_test, scores))
p, r, f, _ = precision_recall_fscore_support(y_test, y_pred, average="binary", zero_division=0)
print("Precision, Recall, F1:", p, r, f)

# save artifacts
joblib.dump(model, "models/autoencoder.joblib")
joblib.dump(scaler, "models/scaler.joblib")

CLI examples

Train:

python src/train.py --data data/transactions.csv --out models/ --epochs 50 --batch-size 64

Evaluate:

python src/evaluate.py --model models/autoencoder.joblib --scaler models/scaler.joblib --data data/transactions_test.csv

Predict:

python src/predict.py --model models/autoencoder.joblib --scaler models/scaler.joblib --input data/new_transactions.csv --output results/predictions.csv

Configuration

Use a config file or CLI flags for:

• model hyperparameters (hidden sizes, epochs, batch size)
• contamination rate (expected anomaly fraction)
• preprocessing options (scaling, feature selection)

Notes & best practices

• Ensure class imbalance handling and realistic contamination setting
• Validate with stratified splits or time-based splits for temporal data
• Monitor false positives—tune contamination and thresholds accordingly

Acknowledgements

Built with PyOD and scikit-learn.