A Predictive model for determining vegetable-derived macronutrients for a diabetic patient

Abstract

Current tools for predicting suitable vegetable plants for consumption by patients of type-II diabetes do not factor for their geolocation. This limits them from accessing and consuming vegetables with the required nutrient content that is needed to control the disease. Given soils have a direct influence on the macronutrient levels found in a plant, there was need to predict optimally the amount of nutrients ingested by the patient, given the geolocation, as soils act as a source of nutrients for vegetable plants. Therefore, this study developed a predictive model that enabled referencing of location-specific nutrients through a georeferenced map of soil macronutrients. The model applied was a hybrid geospatial model that is based on multiple linear regression kriging, that combined a geostatistical and a machine learning technique for predictive mapping, to enhance accuracy of prediction of generated maps. This model was compared against others by support vector machines, multiple linear regression, and regression kriging. However, it performed poorer in map generation compared to the other three models, with the model based on support vector machines providing the highest level of accuracy. The maps generated by this model were then applied in a tool for nutritionists to determine suitable vegetables that is supported by soil at a specific location. This tool will greatly assist in providing tailored dietary plans to type-II diabetic patients.

Keywords: soil macronutrients, human nutrition, hybrid geospatial model, support vector machines, plant determinant tool