HHS Unaccompanied Children Program Operation Dashboard: Guarding Cross-Border Children with Data

This article introduces an operational analysis dashboard for the HHS Unaccompanied Children (UAC) Program built with Streamlit. By integrating machine learning models such as Prophet and XGBoost, it enables operational monitoring, bottleneck analysis, and prediction, supporting efficient management of cross-border child placement services and bridging data with humanitarian care.

The Unaccompanied Children (UAC) Program under the U.S. Department of Health and Human Services (HHS) is responsible for receiving, placing, and caring for minors who enter the U.S. cross-border alone. It faces dual challenges of humanitarian care and administrative management. The traditional approach relying on paper reports and manual statistics is inefficient and makes it difficult to identify systemic operational bottlenecks. This open-source project addresses these pain points by building an interactive analysis dashboard with Streamlit, providing real-time monitoring, predictive analysis, and decision support.

Frontend Framework: The Lightweight Beauty of Streamlit

Streamlit is chosen as the frontend framework. With pure Python code, it allows rapid construction of interactive interfaces without needing HTML/CSS/JS, focusing on business logic and fast iteration.

Visualization and KPI Tracking

Connect to the backend database to display key operational indicators: daily number of children received, placement waiting time, placement site capacity utilization, case processing progress, etc. Dynamic charts support time filtering and drilling down to help identify anomalies.

Bottleneck Analysis

Using process mining technology to analyze the complete path of children from entry to placement, identify the most time-consuming links, and locate weak points such as reception, placement matching, or transportation scheduling.

Multi-Model Integration Prediction

• Prophet: Captures seasonal fluctuations (e.g., increased entry in summer) to generate short-term predictions;
• Random Forest: Handles non-linear relationships, integrating factors like season and policy to predict demand;
• XGBoost: Efficiently processes tabular data and serves as the main predictive model.

Model Validation and Selection Strategy

Compare the performance of different models through backtesting with historical data, automatically select the method suitable for the current scenario, and ensure prediction reliability and interpretability.

• Capacity Planning: Pre-estimate the number of children to receive, coordinate placement site capacity, personnel, and materials;
• Resource Optimization: Adjust resource allocation for bottleneck links (e.g., medical screening delays);
• Emergency Response: Display system capacity in real-time to assist in deciding whether to activate emergency protocols or seek support.

• Data Privacy and Compliance: Must strictly comply with regulations like HIPAA. During deployment, implement data desensitization, access control, and audit logs;
• Model Drift and Update: Establish monitoring mechanisms to trigger model retraining when prediction errors increase;
• Multi-Source Data Integration: Build a unified data pipeline and ETL process to integrate multi-source data from border enforcement, medical care, education, etc.

This project demonstrates how data science can serve the care of vulnerable groups, optimizing processes through prediction, visualization, and bottleneck analysis. For data analysts in the public sector and technical practitioners in social services, it provides a reference for transforming complex social issues into quantifiable optimization problems.