Improving Nutritional Analysis with XGBoost: Predicting Calories from Food Images

Accurately tracking calorie consumption is crucial in today’s health-conscious world. Proper monitoring can help prevent diseases such as diabetes and heart-related conditions. Conventional calorie prediction for food items may require detailed nutritional knowledge about the foods, and before varying, portion sizes can affect accuracy, which contributes to the time-consuming nature and likelihood of errors. This research analyzes the use of XGBoost, an ensemble machine learning technique, which can improve calorie prediction from food images. The XGBoost model, trained by the extracted features from food images, has obtained an R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> score of 0.98, an MAE of 1.359 kcal, and an MSE of 349.2 kcal<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>. Boosting is the main principle behind XGBoost, which is an ensemble based approach that increases accuracy by merging multiple weak learners into strong learners. It gives greater importance to incorrectly predicted instances, allowing the model to concentrate on difficult cases in later iterations. The XGBoost regressor improves this process by using regularization, parallel and by running the model in parallel, efficient tree pruning. XGBoost outperformed conventional models like decision trees, random forests, support vector machines (SVM), and K-nearest neighbors (KNN) in terms of decreasing prediction errors and lowering bias and variance.