The incipient faults classification of transformers is crucial for preventing failures and ensuring reliable operation. This paper presents a machine learning (ML) model designed for transformer fault identification, integrating modified artificial data with a novel 4GM graphical algorithm. The study trains the KNN model on a 4GM-based dataset of 1022 transformers specifically for fault identification. The model’s performance is validated using the IEC TC 10 database, demonstrating effective accuracy in fault classification. Additionally, a modified Artificial Data Generation (ADG) technique is introduced to enhance the dataset, addressing data imbalances and improving model robustness. Initial integration of the ADG technique led to reduced accuracy, necessitating further adjustments. Through iterative modifications, the combination of the 4GM algorithm and the enhanced ADG technique resulted in improved accuracy and robustness. This research will help many researchers in this fields for further improvement of ML performance when data imbalance, low database problem occurred.