Social isolation refers to the health condition in which an individual has no contact with other people, causing clinical risk factors, which combined with various mental health problems increase the mortality and morbidity rate. The detection and prediction of social isolation is essential for treatment and therapy, helping individuals to reconnect with their social environment. For the first stage of the work, RADAR-Base will be used, which is an open source platform for the management and collection of data for the assessment of diseases remotely, using mobile devices and applications. It collects data in two ways such as through a simple configurable questionnaire, with remote notifications, configurable schedules and optional audio capture; and as a passive application that collects sensor data in the background from phones or handheld devices, with real-time transmission, customizable per study and fault-tolerant. The sensors collect several features such as relative location, acceleration, spin, magnetic field, footsteps, light, interaction status, application usage, call log, SMS log, contact list, Bluetooth devices, battery level and local weather. In the next stage, from the features collected by the mobile devices, an exploration of the data will be performed to identify problems such as missing values, outliers and errors in the collected values, managing to identify which are the features that can help the machine learning models to a prediction with high performance. Finally, having the data preprocessed, the Machine Learning models will be built. This process involves selecting the appropriate algorithm, the data for training and evaluating the performance of the models, in order to treat the social distancing in its initial stages and prevent its development.