\n", "\n", " Model Card for Credit Card Fraud Model\n", "\n", "

{ "cells": [ { "cell_type": "code", "execution_count": 1, "id": "67ba05a9", "metadata": {}, "outputs": [], "source": [ "import sys\n", "import json\n", "from IPython import display\n", "import pandas as pd\n", "import numpy as np\n", "import category_encoders as ce\n", "from sklearn.ensemble import RandomForestClassifier\n", "from sklearn.model_selection import train_test_split\n", "from sklearn.metrics import ConfusionMatrixDisplay, RocCurveDisplay, precision_score, recall_score\n", "from sklearn.pipeline import Pipeline\n", "\n", "# parent directory to work with dev\n", "sys.path.insert(0, '..')\n", "import verifyml.model_card_toolkit as mctlib\n", "from verifyml.model_card_toolkit import model_card_pb2, ModelCard\n", "from verifyml.model_card_toolkit.utils.tally_form import tally_form_to_mc\n", "from verifyml.model_tests.utils import plot_to_str\n", "from verifyml.model_tests.FEAT import (\n", " SubgroupDisparity,\n", " MinMaxMetricThreshold,\n", " Perturbation,\n", " SHAPFeatureImportance,\n", " FeatureImportance,\n", " DataShift\n", ")" ] }, { "cell_type": "code", "execution_count": 2, "id": "4a4b10a8", "metadata": {}, "outputs": [], "source": [ "# Credit card fraud Dataset\n", "df = pd.read_csv(\"../data/fraud.csv\")\n", "x = df.drop(\"is_fraud\", axis=1)\n", "y = df[\"is_fraud\"]\n", "\n", "\n", "# Train-Test data Split\n", "x_train, x_test, y_train, y_test = train_test_split(\n", " x, y, test_size=0.5, random_state=50\n", ")\n", "\n", "\n", "## Build ML model with protected attributes as model features\n", "\n", "# Apply one hot encoding to categorical columns (auto-detect object columns) and random forest model in the pipeline\n", "estimator = Pipeline(steps=[('onehot', ce.OneHotEncoder(use_cat_names=True)),\n", " ('classifier', RandomForestClassifier(n_estimators=4, max_features=\"sqrt\", random_state = 882))])\n", "\n", "\n", "# Fit, predict and compute performance metrics\n", "estimator.fit(x_train, y_train)\n", "\n", "output = x_test.copy() # x_test df with output columns, to be appended later\n", "y_pred = estimator.predict(x_test)\n", "y_probas = estimator.predict_proba(x_test)[::, 1]\n", "\n", "precision_train = round(precision_score(y_train, estimator.predict(x_train)),3)\n", "recall_train = round(recall_score(y_train, estimator.predict(x_train)), 3)\n", "precision_test = round(precision_score(y_test, y_pred),3)\n", "recall_test = round(recall_score(y_test, y_pred), 3)\n", "\n", "\n", "# Add output columns to this dataframe, to be used as a input for feat tests\n", "output[\"truth\"] = y_test\n", "output[\"prediction\"] = y_pred\n", "output[\"prediction_probas\"] = y_probas\n", "\n", "\n", "# Dataframe with categorical features encoded\n", "x_train_encoded = estimator[0].transform(x_train)\n", "x_test_encoded = estimator[0].transform(x_test)\n", "\n", "\n", "# Get feature importance values\n", "df_importance = pd.DataFrame(\n", " {\"features\": x_test_encoded.columns, \"value\": estimator[-1].feature_importances_}\n", ")" ] }, { "cell_type": "markdown", "id": "05e72fb8", "metadata": {}, "source": [ "## Get confusion matrix and ROC curve on train/test set" ] }, { "cell_type": "code", "execution_count": 3, "id": "913baf07", "metadata": {}, "outputs": [ { "data": { "image/png": "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\n", "text/plain": [ "

	feature_rank	passed\r", "
age_>=65	10.0	False\r", "
gender_F	13.0	True\r", "
gender_M	16.0	True\r", "
age_40-64	18.0	True\r", "
age_26-39	21.0	True\r", "
age_<=25	26.0	True\r", "

	fpr at current probability threshold	passed\r", "
gender_M	0.009	True\r", "
gender_F	0.024	True\r", "

	fpr at current probability threshold	passed\r", "
age_>=65	0.011	True\r", "
age_26-39	0.021	True\r", "
age_40-64	0.024	True\r", "
age_<=25	0.012	True\r", "

	training_distribution	eval_distribution	p-value	passed\r", "
gender_F	0.526	0.544	0.234	True\r", "
gender_M	0.474	0.456	0.234	True\r", "
age_26-39	0.23	0.229	0.798	True\r", "
age_40-64	0.323	0.321	0.798	True\r", "
age_<=25	0.044	0.05	0.798	True\r", "
age_>=65	0.403	0.4	0.798	True\r", "

	fpr of original data	fpr of perturbed data	ratio	passed\r", "
age_26-39	0.021	0.019	1.14	True\r", "
age_40-64	0.024	0.019	1.268	True\r", "
age_<=25	0.012	0.024	0.482	True\r", "
age_>=65	0.011	0.012	0.959	True\r", "

	age_fpr_max_ratio\r", "
0	2.094\r", "

	gender_fpr_max_ratio\r", "
0	2.574\r", "

	age_fpr_max_ratio\r", "
0	1.175\r", "

	gender_fpr_max_ratio\r", "
0	1.013\r", "

\n", "\n", " Model Card for Credit Card Fraud Model\n", "\n", "

Model Details

Overview

Version

Owners

Regulatory requirements

Considerations

Intended Users

Use Cases

Fairness Considerations

Datasets

Credit Card Dataset

Sensitive data

Sensitive data used in model

Justification

CRM

Sensitive data

Sensitive data used in model

Justification

Quantitative Analysis

Recall - 0.936 (Training Set)

Precision - 0.998 (Training Set)

Recall - 0.77 (Test Set)

Precision - 0.955 (Test Set)

Explainability Analysis

Top 10 most important features

Protected Attributes should not be model's top important features

Shapely Feature Importance Test

Fairness Analysis

Minimum acceptable service

ROC/Min Max Threshold Test

ROC/Min Max Threshold Test

Equal false positive rate

Subgroup Disparity Test

Subgroup Disparity Test

Distribution of attribute subgroups should be silimiar across different datasets

Data Shift Test

Fairness metric for subgroups in original data and perturbed data should be similar

Subgroup Perturbation Test

\n", "\n", " Model Card for Credit Card Fraud Model\n", "\n", "

Model Details

Overview

Version

Owners

Regulatory requirements

Considerations

Intended Users

Use Cases

Fairness Considerations

Datasets

Credit Card Dataset

Sensitive data

Sensitive data used in model

Justification

CRM

Sensitive data

Sensitive data used in model

Justification

Quantitative Analysis

Recall - 0.958 (Training Set)

Precision - 0.981 (Training Set)

Recall - 0.711 (Test Set)

Precision - 0.858 (Test Set)

Explainability Analysis

Top 10 most important features

Fairness Analysis

Minimum acceptable service

ROC/Min Max Threshold Test

ROC/Min Max Threshold Test

Equal false positive rate

Subgroup Disparity Test

Subgroup Disparity Test

Credit Card Fraud Model

Credit Card Fraud Model, without protected attributes as model features

Performance Metrics

Recall - 0.936

Recall - 0.958

Precision - 0.998

Precision - 0.981

Recall - 0.77