Mobile-based food classification for Type-2 Diabetes using nutrient and textual features

Type-2 Diabetes (T2D) is a dreadful disease affecting hundreds of millions of people worldwide, and is linked and worsen by unhealthy lifestyles, especially the poor diet style. However, managing daily diet effectively remains highly challenging for both T2D patients and doctors. In this paper, we proposed, built, and evaluated an effective food classification tool using mobile computing and predictive models to proactively guide T2D patients along their diet selection. This tool provided a comprehensive food database so that patients can conveniently utilize it to record and track their daily diet. More intelligently, the embedded predictive model classified each food item into three classes (e.g., “Choose More Often”, “In Moderate”, and “Choose Less Often”) using its nutrient and textual features. The evaluation results show that it is able to achieve around 93% classification accuracy in the best scenario, which indicates that it is efficient and effective for T2D diet management.