This app harnesses machine learning to predict the presence or absence of diabetes. The app takes it a step further by offering customized diet suggestions tailored to the user's health status, desired diet goals, daily activity volume, and diabetes prediction outcome.
demo video: https://youtu.be/5oK51loKqIQ
This app harnesses machine learning to predict diabetes and provide personalized diet recommendations. Using a Random Forest model, it analyzes user-provided data such as age, gender, family history, blood pressure, height, weight, and glucose levels to accurately predict the presence or absence of diabetes. The app takes it a step further by calculating the user's Total Daily Energy Expenditure (TDEE) based on their BMI, which is derived from their height and weight, as well as their activity level. Utilizing this information, the app offers customized diet suggestions tailored to the user's health status, desired diet goals (cutting, bulking, or standard), daily activity volume, and diabetes prediction outcome. These suggestions consider macronutrient distribution, portion sizes, empowering users to make informed decisions about their dietary habits. With its ability to predict diabetes and provide tailored diet plans, this app serves as a valuable tool for individuals, particularly those with diabetes, in organizing their meals and managing their health effectively.
Diabetes is a common chronic disease in modern society, and it can have severe consequences on the body, especially in old age. This is particularly crucial for individuals already diagnosed with diabetes, as having correct dietary concepts, nutrient proportions, and portion control is vital in managing the progression of the disease. Unfortunately, due to the inconvenience of medical check-ups, many individuals neglect assessing their current risk of developing diseases. Therefore, a system that can quickly utilize certain test data without requiring individuals to visit hospitals would enable them to assess their own risk of disease from the comfort of their homes. This would be extremely important for early treatment and prevention. For many people, there is a lack of understanding regarding nutrient concepts in their diet planning. Often, they consume excessive carbohydrates, inappropriate nutrient proportions, or improper portions, which can greatly contribute to the development of diabetes.
Additionally, for individuals without diabetes, the daily intake of nutrients in proper proportions and quantities is also highly important. Depending on their activity levels, dietary goals, and individual physical condition, the daily nutrient requirements vary among individuals. There is no fixed value, as it is tailored to each person's needs. A well-planned diet can help prevent chronic diseases such as diabetes. Furthermore, correct, and appropriate dietary planning is indispensable for those who aim to build muscle or lose weight. However, busy modern individuals often lack the time and knowledge to extensively study dietary concepts. The prevalence of eating out among the population further contributes to neglecting the content of their meals. Therefore, a user-friendly tool that simplifies the process would allow both diabetic patients and the public to quickly determine how to arrange their diets. Such a tool could potentially play a significant role in overall chronic disease prevention and disease management.
Part1.
The diabetes prediction model:
Diabetes Prediction Model Using the random forest algorithm, I constructed decision trees based on several features (Pregnancies, glucose, blood pressure, skin thickness, insulin, diabetes prediction function, age). The individual decision trees were then combined, and the results were averaged to obtain an overall analysis of whether a person is likely to have diabetes based on the values.
By combining the data from these trees, we ensure obtaining the most accurate predictions. While individual decision trees can provide a result and a narrow range of groups, a forest improves the accuracy of the results through multiple groups and decision processes. It adds randomness to the model by finding the best features within random subsets of features.
For model training, I utilized the random forest algorithm from the scikit-learn library. I used a dataset from Kaggle, consisting of 768 records of body data along with information on the presence or absence of diabetes. This dataset served as the training dataset for the model.
The above figures are the data analysis of the dataset used reveals that most of the data tends to be concentrated within a specific range but also exhibits a continuous distribution. This distribution will serve as a reference for future design considerations.
For example, in the case of the "Diabetes Prediction Function," which relates to family medical history, we can observe that the majority of the data falls below 0.6263, with an average of 0.4719. Therefore, when designing the user interface for the Diabetes Prediction Function dropdown, we can consider the distribution of this data and create options that align with the data patterns.
Part2.
Dietary Planning system:
Based on the user's input of height and weight, the following steps can be taken to calculate the user's Basal Metabolic Rate (BMR) and Total Daily Energy Expenditure (TDEE). This information, along with the diabetes prediction result, can further guide appropriate dietary recommendations. For males: BMR = 66 + (bodyweight * 13.7) + (5 * height) - (6.8 * age) For females: BMR = 655 + (bodyweight * 9.6) + (1.8 * height) - (4.7 * age) Next, the user can select their activity level from a menu, categorized as low, moderate, or high:
Low: Almost no exercise and not involved in labor-intensive work. Moderate: Regular exercise two to three times a week. High: Labor-intensive work, athlete, or exercising more than four times a week.
Based on this activity level categorization, the TDEE can be calculated as follows: Low activity level: TDEE = BMR * 1.2 Moderate activity level: TDEE = BMR * 1.55 High activity level: TDEE = BMR * 1.9
Next, in conjunction with the diabetes prediction result from Part 1. If the prediction indicates a potential risk of diabetes, the system will allocate the user's daily calorie intake directly to the most suitable one for a diabetes patient. The recommended distribution, according to Taiwan's Ministry of Health and Welfare, is to allocate the user's daily calories as follows: protein and fat each account for 25%, and carbohydrates account for 50%.
For users who are predicted to be healthy, they can select their dietary goal (e.g., muscle gain, fat loss, or maintenance). The system will then provide the recommended daily calorie intake based on their goal:
For fat loss: TDEE - 300 kcal For muscle gain: TDEE + 400 kcal For weight maintenance: Intake should be based on TDEE
Once the calorie intake for each nutrient is determined, the corresponding portions of each nutrient can be calculated based on the energy density of each nutrient (carbohydrates: 4 kcal/g, protein: 4 kcal/g, fat: 9 kcal/g).
- The image folder and .csv file should be put under the same folder as the code file.
- The ML training dataset comes from https://www.kaggle.com/datasets/akshaydattatraykhare/diabetes-dataset