Ketogenic diet is often used as diet therapy for certain diseases, among other things, its positive effect related to weight loss is highlighted. Precisely because of the suggestion that KD can help with weight loss, visceral obesity, and appetite control, 100 respondents joined the weight loss program (of which 31% were men and 69% were women). The aforementioned respondents were interviewed in order to determine their eating habits, the amount of food consumed, and the time when they consume meals. Basic anthropometric data (body height, body mass, chest, waist, hips, biceps, and thigh circumferences) were also collected, in order to be able to monitor their progress during the different phases of the ketogenic diet. Important information is the expected body mass during the time frame of a certain keto diet phase. This information is important for the nutritionist, medical doctor, as well as for the participant in the reduced diet program; therefore, the model was developed that modified the original equation according to Wishnofsky. The results show that women lost an average of 22.7 kg (average number of days in the program 79.5), and for men the average weight loss was slightly higher, 29.7 kg (with an average of 76.8 days in the program). The prediction of expected body mass by the modified Wishnofsky’s equation was extremely well aligned with the experimental values, as shown by the Bland-Altman graph (bias for women 0.021 kg and −0.697 kg for men) and the coefficient of determination of 0.9903. The modification of the Wishnofsky equation further shed light on the importance of controlled energy reduction during the dietetic options of the ketogenic diet.

The use of mathematical modeling and optimization in nutrition with the help of artificial intelligence is indeed a trendy and promising approach to data processing. With the ever-increasing amount of data being generated in the field of nutrition, it has become necessary to develop new tools and techniques to help process and analyze these data. The paper presents a study on the development of a neural-networks-based model to investigate parameters related to obesity and predict participants’ health outcomes. Improvement techniques of model performances are made (classification performance by reducing overfitting, capturing non-linear relationships, and optimizing the learning process). Predictions are also made with the random forest model to compare the performance of accuracy and prediction scores of two different models. The dataset contains data relating to the obesity of 200 participants in a weight loss program. Information is collected on their basic anthropometric data, as well as biochemical data, which are significant parameters closely related to obesity. It is important to note that weight loss is not always linear and can vary based on individual factors; so, a prediction is made on supervised learning based on patient data (before the diet regime, during the regime, and reaching the desired weight). The dataset is trained on individuals features such as age; gender; body mass index; and biochemical attributes such as MCHC (Mean Corpuscular Hemoglobin Concentration), cholesterol, glucose, platelets, leukocytes, ALT (alanine aminotransferase), triglycerides, TSH (thyroid stimulating hormone), and magnesium. The results of the developed neural network model show high accuracy, low loss in training, high-precision predictions during evaluation of the model, and improved performance over other machine learning models. Calculations are conducted in Anaconda/Python. Overall, the combination of mathematical modeling, optimization, and AI offers a powerful set of tools for analyzing and processing nutrition data. As our understanding of the relationship between diet and health continues to evolve, these techniques will become increasingly important for developing personalized dietary recommendations and optimizing population-level dietary guidelines.