A Comparison of Personalised Recipe Recommendation System Machine Learning Models

With that due to the rise in lifestyle related dise aeses and the consumers’ growing interest in personalized nutrition, there is a need for intelligent recipe recommendation systems that dynamically adapt to users’ health goals and ingredient available. Taking a deeper look at this, this work has the aim of designing and conducting performance evaluation of a personalized recipe recommendation system based on the nutritional information and machine learning methodologies. In the proposed system, user provided numerical nutritional targets (calories, protein, fat, fiber, sodium, and cholesterol) and free text ingredient lists are accepted to produce customized recipe suggestions. A comparison of four machine learning support vector machines (SVM) algorithms, transformers, k nearest neighbor (KNN), and long short term memory (LSTM) networks shows that KNN and SVM offer just a little bit of predictive accuracy in regards to static contexts; however, they do not appropriately model the complex dependencies between ingredients and nutritional attributes. A transformer can outperform conventional models via its attention mechanism which comes with the ability to model ingredient relationships and nutritional constraints, but is computationally expensive and therefore is not suitable for real time applications. However, Transformers offer a desirable trade off: long term dependency capture at a usable computational efficiency, and thus are a good fit for real time recipe personalization. As shown by this research, the selection of an algorithm such as a dynamic memory or memory networks, which condition prediction on item context, is of high importance for recommendation systems among health conscious users due to the benefit that future systems of recommendation will derive from hybrid models based on contextual attention in conjunction with sequential learning. The next generation of intelligent culinary assistants will be such systems, and it will enable dietary recommendations that target health optimization by feeding personalized ones that are simultaneous predictions and are relevant to nutritional as well as real-time responsive needs.