Edge computing allows for real-time intelligent services to be provided close to where data is generated, but edge devices have limited computing power. A continuous deep learning approach based on knowledge transfer was used to reduce training time without sharing data. Each edge device fine-tunes a deep learning model with its local dataset, then transfers the updated model to the next device. By emulating multiple edge devices and using Convolutional Neural Networks for classification, the prediction accuracy improved with each iteration of model transfer in the edge computing environment.