Non-invasive glucose monitoring is a critical area of research in the management of diabetes mellitus. Traditional invasive methods for blood glucose testing, although accurate, present significant drawbacks such as discomfort and risk of infection. This study explores the use of Raman spectroscopy combined with machine learning algorithms for the prediction of glucose levels, offering a potential non-invasive alternative. We simulate glucose levels using diluted glucose solutions and apply Raman spectroscopy to collect spectral data. The data is then processed using an Improved Modified Polynomial (IMP) method to correct background noise, followed by classification using Support Vector Machine (SVM) and Extra Trees algorithms. Our experimental results demonstrate that the processed data significantly improves the accuracy of glucose level predictions, with the SVM model achieving an average accuracy of 81.2% and the Extra Trees model reaching 83.7% when using a 12th-order polynomial for baseline correction. These findings suggest that Raman spectroscopy, when coupled with advanced data processing and machine learning techniques, holds promise for developing a reliable non-invasive glucose monitoring systems.