Pogi: A Machine Learning Pipeline for Predicting U.S. Federal Fiscal Balances Using Generative AI

Pogi is an open-source machine learning pipeline dedicated to predicting and analyzing the fiscal account balances of the U.S. federal government. By combining generative AI with traditional statistical learning methods, it addresses the limitations of traditional budget analysis that relies on manual experience and static reports. It provides data-driven insights for fiscal analysts and policymakers, supporting core scenarios such as budget execution prediction, congressional appropriation scenario testing, and audit anomaly detection.

U.S. federal financial management involves thousands of fiscal accounts, each with unique appropriation cycles, expenditure patterns, and balance change rules. According to the requirements of 31 U.S.C. 1511-1514, the President must review federal expenditures at least four times a year. The SF 133 Budget Execution Report records historical data from 1998 to the present, but the value of this massive data has not been fully exploited.

Pogi's core objective is to build an automated intelligent analysis pipeline that learns patterns from historical financial data, predicts future account balance changes, identifies abnormal fund flows, improves the accuracy of budget preparation, and provides an early warning mechanism for audits.

Pogi adopts a modular design and integrates multiple machine learning algorithms:

• Regression tasks: Covers linear regression, Ridge/Lasso regularization, decision trees, random forests, gradient boosting, support vector regression, and multi-layer perceptron neural networks;
• Classification tasks: Includes perceptrons, logistic regression, random forests, AdaBoost, and gradient boosting classifiers;
• Data preprocessing: Built-in processes for missing value handling, feature scaling, polynomial expansion, dimensionality reduction, etc., to ensure input data quality.

Pogi integrates two core official data sources:

1. SF 133 Budget Execution Report: Published by the White House Office of Management and Budget (OMB), linked to the GTAS system, containing budget resources, obligations, and expenditure data of federal agencies;
2. File A data from USAspending.gov: Published monthly as required by the DATA Act, covering budget resources, obligations, and expenditure information for all fiscal accounts.

The project builds a decades-long historical data warehouse, supporting data import in Excel/CSV formats to provide sufficient learning samples for the models.

Core application scenarios:

1. Budget execution prediction: Analyze historical expenditure patterns to predict future account balance changes and assist in fund planning;
2. Congressional appropriation scenario testing: Simulate the impact of different appropriation plans on account balances and provide quantitative decision-making basis;
3. Audit preparation and anomaly detection: Identify fund flows that deviate from normal patterns and warn of compliance risks.

Value: Improve financial analysis efficiency, introduce data-driven decisions to reduce subjective bias; open-source nature promotes transparency and standardization in public financial management.

Pogi provides multiple usage scenarios:

• Cloud quick start: Google Colab notebooks, no local configuration required; run by uploading data or mounting Drive;
• Local deployment: Install dependencies via pip and run in Jupyter Notebook;
• Interactive web application: Streamlit-based graphical interface supporting data upload, model configuration, training, and result viewing.

All model training and evaluation are completed locally to ensure the security of sensitive financial data.

Pogi's innovations include:

1. Introducing the concept of generative AI into the field of public fiscal prediction;
2. Complete data science workflow: Covers data ingestion, cleaning, feature engineering, model training, evaluation, and visualization;
3. Interpretability support: Provides visualization tools such as scatter plots, residual analysis, ROC curves, and confusion matrices;
4. Built-in statistical analysis functions: Descriptive statistics (mean, median, etc.), distribution characteristics (standard deviation, skewness, etc.), hypothesis testing (t-test, ANOVA, etc.).

Pogi demonstrates the great potential of machine learning in the field of public financial management. By combining advanced algorithms with government financial data, it provides powerful tools for budget analysis, appropriation decision-making, and audit supervision. With project development and community contributions, it is expected to become a benchmark case for data science practices in the public sector, promoting more government agencies to adopt data-driven decisions.

For readers concerned about government transparency, fiscal efficiency, and technological innovation, Pogi is both a practical tool and a research example of artificial intelligence serving the public interest.