Step-by-Step: Building Financial ML Models with Vertex AI and the What-If Tool

Learn how to create and deploy financial machine learning models using Vertex AI and the What-If Tool with our comprehensive guide.

Introduction

In the rapidly evolving financial sector, leveraging artificial intelligence (AI) for financial modeling is becoming indispensable. Traditional methods, often reliant on manual spreadsheet management, are no longer sufficient to handle the complexity and volume of modern financial data. This is where AI Financial Modeling steps in, revolutionizing the way financial projections and analytics are performed. In this guide, we will walk you through building and deploying financial machine learning (ML) models using Vertex AI and the What-If Tool, empowering you to make data-driven decisions with confidence.

Why AI Financial Modeling?

Financial professionals today grapple with vast datasets and the need for precise, timely insights. Manual data entry and analysis not only consume significant time but also increase the risk of errors, leading to potential revenue loss and delayed decision-making. AI Financial Modeling addresses these challenges by automating data processing, enhancing accuracy, and providing dynamic visualizations that transform raw data into actionable insights.

Benefits of AI-Driven Models

• Efficiency: Automate repetitive tasks, reducing the time spent on data entry and analysis by up to 90%.
• Accuracy: Minimize errors associated with manual processes, ensuring reliable financial projections.
• Scalability: Easily handle increasing data volumes and complexities without compromising performance.
• Insightful Visualizations: Generate dynamic charts and reports that facilitate better understanding and decision-making.

Getting Started with Vertex AI

Vertex AI is Google Cloud’s unified platform for building, deploying, and scaling ML models. It integrates various ML offerings into a seamless development experience, making it an ideal choice for financial modeling.

Setting Up Your Environment

To begin, ensure you have a Google Cloud Platform (GCP) project with billing enabled. Follow these steps to set up your environment:

1. Enable the Compute Engine API: Navigate to the Compute Engine section of your Cloud Console and enable the API.
2. Enable the Vertex AI API: Go to the Vertex section and enable the Vertex AI API.
3. Create a Notebooks Instance: From the Vertex AI section, create a new TensorFlow Enterprise 2.3 instance without GPUs and open JupyterLab.
4. Install XGBoost: In your JupyterLab terminal, install the latest version of XGBoost supported by Vertex AI using the command:
   bash
pip3 install xgboost==1.2

Building Your Financial ML Model

Step 1: Data Preparation

Using a preprocessed mortgage dataset, we’ll train an XGBoost model to predict loan approvals.

1. Download the Dataset:
   bash
gsutil cp 'gs://mortgage_dataset_files/mortgage-small.csv' .
2. Load and Shuffle Data:
   “`python
   import pandas as pd
   from sklearn.utils import shuffle

data = pd.readcsv(‘mortgage-small.csv’, dtype=COLUMNNAMES).dropna()
data = shuffle(data, randomstate=2)
labels = data[‘approved’].values
data = data.drop(columns=[‘approved’])
3. **Encode Categorical Variables**:python
data = pd.getdummies(data, columns=dummycolumns)
4. **Train-Test Split**:python
from sklearn.modelselection import traintestsplit

xtrain, xtest, ytrain, ytest = traintestsplit(data.values, labels)
“`

Step 2: Training the XGBoost Model

Train the model using the training dataset:

import xgboost as xgb
from sklearn.metrics import accuracy_score

model = xgb.XGBClassifier(objective='reg:logistic')
model.fit(x_train, y_train)
y_pred = model.predict(x_test)
accuracy = accuracy_score(y_test, y_pred.round())
print(f"Model Accuracy: {accuracy * 100:.2f}%")

Expected accuracy: ~87%

Step 3: Interpreting the Model with the What-If Tool

The What-If Tool allows you to visualize and analyze your model’s predictions.

1. Create Visualization:
   “`python
   from witwidget.notebook.visualization import WitWidget, WitConfigBuilder

configbuilder = (WitConfigBuilder(testexamples.tolist(), data.columns.tolist() + [‘mortgagestatus’])
.setcustompredictfn(model.predictproba)
.settargetfeature(‘mortgagestatus’)
.setlabelvocab([‘denied’, ‘approved’]))
WitWidget(config_builder, height=800)
“`
2. Explore Data Points: Modify feature values to observe changes in predictions.
3. Perform Counterfactual Analysis: Identify similar data points with opposite predictions.
4. Analyze Feature Impact: Use partial dependence plots to understand feature influences.
5. Assess Performance and Fairness: Review confusion matrices and fairness metrics.
6. Review Feature Distribution: Ensure balanced datasets for accurate modeling.

Deploying Your Model with Vertex AI

Deploy your trained model to Vertex AI for scalable and accessible predictions.

Deployment Steps

1. Create a Cloud Storage Bucket:
   bash
gsutil mb -l us-central1 gs://your_storage_bucket_name
2. Upload the Model:
   bash
gsutil cp ./model.bst gs://your_storage_bucket_name
3. Upload the Model to Vertex AI:
   bash
gcloud beta ai models upload --display-name=xgb_mortgage --artifact-uri=gs://your_storage_bucket_name --container-image-uri=us-docker.pkg.dev/cloud-aiplatform/prediction/xgboost-cpu.1-2:latest --region=us-central1
4. Create and Deploy to an Endpoint:
   bash
gcloud beta ai endpoints create --display-name=xgb_mortgage_v1 --region=us-central1
gcloud beta ai endpoints deploy-model your_endpoint_id --region=us-central1 --model=your_model_id --display-name=xgb_mortgage_v1 --machine-type=n1-standard-2 --traffic-split=0=100
5. Test the Deployed Model:
   Create a predictions.json file and use the gcloud command to make predictions.

Cleanup

To avoid unnecessary costs, stop or delete your notebook instance and other resources when not in use.

Conclusion

Building and deploying financial ML models with Vertex AI and the What-If Tool enables financial professionals to harness the full potential of their data. By automating complex tasks and providing insightful visualizations, AI-driven financial modeling transforms analysts into strategic decision-makers, enhancing productivity and accuracy.

Ready to revolutionize your financial analysis with cutting-edge AI tools? Get started with Reef today!