Early identification of potentially low performing community health workers using an ensemble classification model

Purpose - Community health workers (CHWs) are vital to addressing public health system limitations in developing countries. However, effective identification and support of underperforming CHWs remains a challenge. This study develops a predictive model to proactively identify underperforming CHWs, facilitating targeted interventions for improved CHW programmes. Design/methodology/approach - We developed a predictive model to identify underperforming CHWs in Uttar Pradesh, India. Data from 140,101 CHWs over a 12-month period was used to build, test and validate the model. Classification techniques, ensemble modeling and a model tuning algorithm were employed for accuracy optimization and early identification. Findings - Logistic regression, decision trees and random forests yielded the best performance. While ensemble models offered no significant performance improvements over the base models, the model tuning algorithm effectively increased prediction accuracy by 19 percentage points. This enabled early identification of poor-performing CHWs and high-risk CHW clusters early in the year. Practical implications - The developed model has significant potential to improve CHW programmes. It enables targeted support, feedback and resource allocation, leading to enhanced CHW performance, motivation and healthcare outcomes in the communities they serve. The model can provide personalised feedback to help CHWs overcome challenges and dynamic clustering facilitates proactive identification and tailored support for those at risk of underperformance. Originality/value - This study is the first attempt to use predictive modelling to identify underperforming CHWs, advancing the nascent field of CHW performance analytics. It underscores the effectiveness of digital technologies and data in improving CHW programmes.