In the past months, many countries have adopted varying degrees of lockdown restrictions to control the spread of the COVID-19 virus. According to the existing literature, some consequences of lockdown restrictions on people’s lives are beginning to emerge. To inform policies for the current and/or future pandemics, particularly those involving lockdown restrictions, this study adopted a data-driven Machine Learning approach to uncover the short-term effects of lockdown on people’s physical and mental health. An online questionnaire launched on 17 April 2020 was completed by 2,276 people from 66 countries. Focusing on the UK sample (N=382), 10 aggregated variables representing participant’s living environment, physical and mental health were used to train a RandomForest model to predict the week of survey completion. Using an index of importance to identify the best predictor among the 10 variables, self perceived loneliness was identified as the most influential variable. Subsequent statistical analysis showed a significant U-shaped curve for loneliness levels, with a decrease during the 4th and 5th lockdown weeks. The same pattern was replicated in the Greek sample (N = 129). This suggests that for the very first period of time, the adopted lockdown measures affected people’s evaluation of their social support leading to a decreased sense of loneliness.