This paper investigates the possibility of predicting physical weakening in older adults in view to detect early the on-set of frailty process. This study is based on two types of features, namely (i) measured features, which are calculated objectively using performance tests and questionnaires, and (ii) self-reported features, which are based on the older person's auto-evaluation. Two machine learning-based models are proposed. The first one identifies the potential occurrence of physical weakening by comparing the evolution of the aforementioned features between two time slots. The second one predicts a future worsening based on the current values of these features. Both models are evaluated and interpreted using a public dataset. © 2021 IEEE.