In developing countries like India, constant monitoring of the health conditions of transformers with a cloud-based IoT approach is not feasible. Most of the transformers are located in remote areas, therefore, a cloud-based IoT approach needs an uninterrupted Internet connection to run the complex or machine learning algorithms. However, edge computing brings the power of the cloud to the device. In this paper, we proposed an IoT edge computing approach for constant monitoring of transformers using humming sound and machine learning methods. We have collected the humming sound data of normal and faulty categories, then applied the windowing approach to segment humming sound into 2 seconds segments. Mel frequency cepstral coefficients (MFCC) is used to extract features from the sound and the dataset is trained using an SVM classifier. SVM with polynomial kernel performs better and gives 98.77 % average cross-validation accuracy. The trained model is optimized according to the IoT edge computing hardware configuration and deployed to the device for constant health monitoring. The device constantly recording the humming sound and classify each 2 seconds segment into the normal or faulty category and sending the insights to the cloud at regular interval.