A Novel Two-Step Semi-Supervised Learning Algorithm for Discovering Participant Eligibility for the Affordable Connectivity Program

The Affordable Connectivity Program (ACP) is designed to provide financial support for broadband services and devices, ensuring low-income households can stay connected with minimal expense. Despite its benefits, a significant number of eligible households are unaware of their potential to receive ACP benefits. To bridge this gap and alleviate financial burdens, especially for those with low or no income, a robust machine learning approach is essential for predicting ACP eligibility. This study faced several challenges, including a dataset with numerous features and null values, necessitating thorough data cleaning and feature selection. Additionally, the presence of a substantial amount of unlabeled data called for a highly robust semi-supervised model. We introduce a novel two-step semi-supervised learning algorithm tailored for this purpose. Initially, a classification model is trained on labeled data. Subsequently, a model-either classification or clustering-is employed to label the unlabeled data. The newly labeled data is then integrated into the existing labeled dataset, further refining the initial classification model. This enhanced model is ultimately utilized to determine participant eligibility for the ACP. Our proposed semi-supervised algorithm, integrating both classification and clustering models, achieves an impressive F1 score of 0.9980, demonstrating its effectiveness in identifying eligible participants for the ACP program.