Prediction of loneliness in older people

Older people are especially vulnerable to loneliness that has become a major public health concern for people in later life. In this paper, we propose a machine learning based approach to predict loneliness probability using two gradient boosting algorithms, XGBoost and LightGBM. The predictive models are built on a large nationally representative sample from the English Longitudinal Study of Ageing (ELSA) with 7 successive waves (2002 ~2015). The system achieves good performance with a high AUC (Area Under Curve) of over 0.88 and a low LogLoss (Logarithmic loss) of 0.22 on the test data by both algorithms. Moreover, we investigate a wide range of variables to identify significant risk factors associated with loneliness. Specific categories associated with important variables are also recognized by the model. Such information will further enrich our understanding and knowledge of loneliness causes in the elder.