M.S. Applied Data Science - Capstone Chronicles 2025

24

functionality. The primary “make prediction” tab presents a structured input form where users can specify critical parameters including product type (spanning biologics, devices, drugs, food/cosmetics, tobacco, and veterinary categories), recall status, distribution pattern details, specific reason for recall, center classification date, and geographical information regarding the recalling firm. These carefully selected parameters form the foundation for the machine learning model’s predictive analysis of recall classification outcomes. In the “model information” section, users gain valuable insights into the underlying predictive model through comprehensive performance visualizations. The interface displays detailed metrics across various classification categories, presenting precision, recall, and F1-scores to provide transparency regarding the model’s reliability. Additionally, a feature importance graph illustrates the relative influence of different

input parameters on classification decisions, offering users deeper understanding of the factors driving recall categorization and enhancing interpretability of prediction results. For organizations managing multiple recall scenarios, the “Batch Processing” functionality offers significant operational efficiency by accepting CSV file uploads containing multiple recall scenarios. This streamlined approach requires the same structured data as individual predictions—product type, status, distribution pattern, center classification date, recall reasons, and firm location information—but enables simultaneous processing of numerous cases, delivering comprehensive classification predictions that can inform large-scale quality assurance initiatives and regulatory compliance strategies. Link: https://ads599-recall-classification.streamlit.app/

28