We present a physics-based theory and its computational simulation on the performance of a prospective nanodevice based on Machine Learning inside of an Internet of Bio-Nano Things networking to carry out the optimization of artificial incoming of Calcium 2+ ions in beta cells aimed to push out granules of insulin. Although the engineered nanodevices can carry out an optimal task, the inclusion of a Machine Learning algorithm would guarantee the wellness as to avoid scenarios where nanodevices might be running with highest systematic errors.