Diabetic patients need to track and manage their blood glucose level (BGL) and forecast its future readings to have a successful diabetes control. Following the suggested steps beforehand can prevent high or low BGL. Also, recent scientific and practical health-related research studies have demonstrated encouraging outcomes when using machine learning (ML) approaches and techniques, particularly for BGL monitoring and forecasting. The present study aims to simulate and apply ML techniques for glucose forecasting in diabetes patients. Specifically, long short-term memory (LSTM) networks will be used to tackle the problem of predicting BGL and active insulin levels. The work builds upon a research of ML approaches in the context of Diabetes Mellitus, and includes the creation of the predictive model and a web application for testing and exploring the forecasting model’s results. The results produced with LSTM demonstrated superior performance. The LSTM with 2 Layers model outperformed other models, including a multilayer perceptron and Ensemble models which are combinations of various models (Random Forest (RF), Support Vector Machines (SVM) and K-nearest neighbors (KNN)) to improve overall performance. The performance was evaluated mainly using the Mean Squared Error (MSE) and the Root Mean Squared Error (RMSE) as metrics. The LSTM with 2 Layers model stands out as the top performer based on its MSE. Scoring the lowest MSE of 14,43% indicates a superior fit to the data compared to other models. LSTM with 2 Layers delivers the most accurate predictions on average. This is evident from its impressive RMSE of 37.98%, signifying that the model’s errors are generally closer to zero. On the other hand, combined models such as RF, SVM and KNN have an MSE and RMSE of 19.48% and 44.13%, respectively. These findings demonstrate the effectiveness of LSTM in accurately predicting changes in glucose levels, providing a solid foundation for future developments.