As machine learning (ML) adoption grows, small and medium enterprises face challenges in using complex models while maintaining sustainability goals. Current research has mainly focused on sustainability in training deep learning models like artificial neural networks, but little work has addressed sustainable development of ML models, particularly random forest (RF) models. To fill this gap, this study proposes a sustainable training design demonstrated by RF models as design science research. The approach is demonstrated through a case study in a data matching company. Key aspects for environmentally sustainable training include optimised feature selection, hyperparameter tuning and incremental learning, which reduce energy consumption and computing resources. Economic sustainability is achieved by enhancing workflows with accurate models and enabling low entry barriers for companies. Social sustainability is supported by involving employees in model development and offering practical workshops, fostering engagement and smooth integration into business processes.