Previous studies have examined the effects of hyperparameters on random forests, but little research has been done in the context of fall detection. To address this gap, our study aimed to examine how hyperparameters influence the performance and training time of a random forest algorithm used in fall detection systems. Our findings highlighted the best range of values for each hyperparameter to achieve high performance. Moreover, we discovered that certain combinations of hyperparameters could either enhance or reduce the random forest’s performance compared to the default settings. To conduct these investigations, we performed experiments using two datasets: MobiAct v2.0 and UP-Fall, which were collected from accelerometers in smartphones and wearables. These insights can contribute to the optimization of hyperparameters for more effective fall detection systems