ADS Capstone Chronicles Revised

26

4.4.1 Preparation To make this model patient specific, an additional score is calculated that extends the original score based on general recommendations to include patient demographics and conditions. This personalized score gives an individual better insight into which meals are best for their circumstances. The weights for the patient-specific score is calculated by averaging the base weights with the adjusted weights of applicable health conditions, shown in Table 3.

menu items or individual foods are ‘recommended’ or ‘not recommended’ based on the patient scores created and the nutritional value of the items. The classification models trained and tested on the menu and individual food data are Logistic Regression, Random Forest, XGBoost, and Neural Networks. With Support Vector Classifier again only for the individual foods dataset. This was done this way because Support Vectors work better with smaller dataframes. These classification methods are used in a step toward user-friendly recommendation application, as the binary output is more easily interpreted than that of the continuous, numeric patient score.

Table 3 Weight of Nutritional Component based on Condition

= _ ℎ *(1− | − 52.2| 52.5 ) + _ ℎ * (1 − 7 . 5 ) + _ ℎ * ( 7 . 5 ) + _ ℎ * ( 2 0 ) + _ ℎ * (1 − 7 6 5 ) (2) + _ ℎ * (1 − 2 5 )

Decisions on weights were made by putting more emphasis on the nutritional components that are more important to monitor with the given health condition. Following the averaging of the weights, the patient-specific score is calculated similarly to the general score, with the addition of a sodium and fat aspect.

228