Diabetes is a persistent condition characterized by the body’s inability to properly utilize insulin. Therefore, noninvasive methods have been proposed to mitigate the pain and risk of infection due to non-required blood extraction compared to the methods for glucose measurement. The invasive methods are done by finger-pricking blood samples from the body. However, this method increases the risk of blood-related infections and pains. Hence, non-invasive Glucose in the blood is an essential method to measure glucose levels. This study successfully implemented a portable embedded system device based on a multi-spectral sensor and Raspberry Pi 4 microprocessor for noninvasive glucose measurement. This study used machine learning, including Multiple Linear Regression and Support Vector Machine, to predict the glucose level with over 90% accuracy. Moreover, the Clarke error grid analysis was used to analyze the accuracy results of this study. It also compared with invasive measurements and previous studies.