The utilisation of rotating machinery across diverse industrial applications underscores the critical importance of evaluating its reliability to ensure the safe operation of these systems. Bearings, as fundamental components within engineering facilities, hold particular significance; their malfunction can result in severe safety incidents, heightened maintenance expenditures, and considerable economic consequences. Extreme learning machine (ELM) represents a machine learning approach that proficiently addresses numerous challenges inherent in conventional machine learning algorithms. Nonetheless, the efficacy of the ELM may deteriorate and yield inaccurate results due to an inappropriate use of its parameters, which include input weights, biases, and the number of hidden neurons. This paper proposes an improved ELM (IELM) model that incorporates the Hippopotamus optimization algorithm (HOA) to optimise the parameters and enhance the performance of the ELM in rotating machinery reliability assessment. Initially, the HOA method is employed to identify optimised parameter values for the ELM. Subsequently, these optimised values are integrated into the proposed IELM-HOA framework for the purpose of fault classification. This study utilises time-domain statistical features to extract significant information from the vibration signals. The dataset comprises vibration signals derived from both online bearing datasets and experimental bearing datasets. The findings indicate that the proposed IELM-HOA method enhances the performance of the ELM technique. Furthermore, it demonstrates the capability to exceed and compete with recently introduced fault diagnosis methodologies