Dropout prediction in home care training

In Washington state (WA), SEIU 775 Benefits Group provides basic home care training to new students who will deliver care and support to older adults and people with disabilities, helping them with self-care and everyday tasks. Should a student fail to complete their required training, it leads to a break in service, which can result in costly negative health outcomes (e.g. emergency rooms and hospitalization) for their clients [1]. In this paper we describe the results of utilizing machine learning predictive models to accurately identify students who exhibit a higher risk of drop out in two areas: (1) dropping out before attending first class[first class attendance]; and (2) dropping out before completing the training[training completion]. Our experimental results show that AdaBoost algorithm gives a useful result with ROC_AUC = 0.627±0.013 and Precision at 10 = 0.73±0.12 for first class attendance and ROC_AUC = 0.680±0.024 and Precision at 10 = 0.67±0.20 for training completion without relying on additional assessment data about students. In addition, we demonstrate the use case for constructing larger decision trees to help front-line training operations staff identify intervention strategies that create the most impact in preventing dropout.