Predicting and Understanding Care Levels of Elderly People with Machine Learning

Abstract

Home care is a particularly important service for elderly people who need assistance with their management of everyday life from both formal and informal caregivers. In Europe 80% of care is provided by family and friends. Therefore, strengthening the home care by informal caregivers is a crucial task and requires smart solutions to stabilize the current situation. INGE smart solution addresses the challenge of determining and predicting accurate and appropriate care levels for the growing population of elderly individuals in need of care. To be able to achieve this goal, information and structured data are needed to support the smart approach of service development effectively. The proposed approach in INGE utilizes assessment ratings from consultancy visits as features to train and test the developed machine learning model. The performance of the developed random forest-based machine learning model is evaluated using various metrics such as accuracy, precision, recall, F1-score, and confusion matrix and compared with real data collected during the INGE project using the INGE app. The proposed approach achieves an 80% accuracy rate in predicting the care level of care dependents based on category ratings. The used dataset for training the model consists of 454 consultancy visits. The trained model shows that ratings of category self-care have the highest impact on care dependent’s care level sorting as it contributes by 27.1%. This study shows the potential of the INGE smart solution in optimizing both home care situation and care level classification.