Rotating machines are critical assets across various industries, playing a fundamental role in power transmission, energy conversion, and fluid motion. They underpin numerous engineering applications, making their reliability essential. Faults in rotating machines can result in substantial financial losses and, in severe cases, pose risks to human life, emphasizing the need for effective fault prevention strategies. Motivated by the pivotal role of machine learning in this area, this study explores the potential of the Light Gradient Boosting Machine for fault identification in rotating machines. The research focuses on addressing data imbalance issues, analyzing typical features in both time and frequency domains, and evaluating sensing methods. The primary objective was to assess the feasibility of accurately and robustly identifying isolated faults - such as unbalance, misalignment, and looseness. This analysis was conducted using a pre-existing dataset of rotating machine faults, incorporating a limited set of vibration signals alongside classical vibration features. The study demonstrated that LightGBM can be successfully used to classify rotating machinery fault types in both the time and frequency domain, achieving overall accuracies of over 89% using only one classical statistical feature and over 94% with three, even in the presence of data imbalance.