Nowadays, monitoring systems are used to ensure elderly people are safe at home. New approaches consist of developing automatic activity monitoring systems in smart homes equipped with wearable sensors, environmental sensors and visual sensors. In this framework, we proposed an approach that provides an activity recognition system for elderly people in a smart home. Our proposed system is built on three phases. The first phase consists of integrating the quality of context (QoC) of data from the IoT environment. The second phase classifies and labels the sensors based on the QoC using the Fuzzy FCA technique. Finally, the activity recognition phase is based on machine learning by applying Random Forest (RF) with hyperparameter optimization. Following the implementation of our proposed system, the application of RF with hyperparameter optimization demonstrated impressive results. Achieving a precision of 98% these outcomes surpass the performance of RF without optimization. This highlights the efficacy of our approach in enhancing the accuracy and reliability of activity recognition for elderly individuals in smart home environments.