BACKGROUND

Diabetes is an ongoing non-transferable illness that requires everyday prescription and close observing. Self-management is important to stay away from complexities, and great information is expected to oversee self-management appropriately. Currently, more than one in ten people worldwide have diabetes. Additionally, a list of countries where one in five people with diabetes is growing. Since the initial publication in 2000, the absolute prevalence of diabetes mellitus among adults aged 20 to 79 has increased from 151 million people, or 4.6% of the world's population, to 537 million people, or 10.5% of the current global population.

The use of information technology has permeated every aspect of our lives. Online resources are expanding quickly. It is not feasible to bring along books and notes on various diseases in the present era. Nearly everyone has a mobile phone these days, and many of them are smartphones and tablets that can run a variety of apps and games. However, there aren't many apps out there that are made with scientific foundation, easy to use user interfaces for diabetes management and made for specific countries or regions so that people can use it without much prior knowledge. If we speak about Sri Lanka, there is hardly any application available which supports both Sinhala and English.

DiaBeta system is a mobile application which support both android and iOS, registered users can predict whether they have prediabetes or diabetes type 2. System can recommend meal plan which contains Sri Lankan cuisines and exercises plan for the users according to their preferences, users can identify whether they have any health risks. Other than that, users can measure their blood glucose level non-invasively and can measure heart rate and blood oxygen level as well via our DiaBeta hardware device. We hope to add our DiaBeta mobile application to Google play store and introduce the application for the patients in local diabetes clinics in Sri Lanka. By delivering notifications to the user's account about important announcements, easily created quizzes, and news, this study aims to determine the degree of knowledge, attitude, and commitment among individuals. The digital logbook feature is an additional feature that enables users to add, edit, and delete information about their daily blood sugar, insulin, and medication intake. Users can also generate reports from their logged data, receive reminders about when to take their medications, clinic dates, and other crucial information, and educate others about diabetes. Finally, this application will offer the public a very effective and productive service by utilizing all these services

RESEARCH PROBLEM

Life with diabetes is becoming increasingly common as people's standards of living grow. Because of this, it's crucial to be able to promptly and effectively identify and assess diabetes. Initially, type 2 diabetics often show no symptoms. For many years, they may not exhibit any symptoms. Therefore, diabetes prediction is crucial, and it's quite helpful if users may utilize a smartphone application to do so. Even though there are numerous diabetes management apps available in the Apple and Google app stores, the majority of them do not provide a mechanism to more correctly forecast a user's diabetes conditions using machine learning methods.

Both heart rate and blood sugar levels are now measured using pricey equipment. Particularly scarce and costly are the test strips used to monitor blood glucose levels. Additionally, hand punctures are often seen as an aggressive method of determining blood glucose levels. Some folks are not content to only have blood glucose levels measured and their palms pierced. The answer is to use Arduino Uno to implement a low-cost monitoring system that measures patients' blood glucose levels and heart rates by placing sensors in close proximity to the human body as a non-invasive method of measuring blood glucose levels and heart rates. It also connects via Bluetooth to the DiaBeta mobile application.Diabetes is increasingly common in people's everyday lives as their level of life grows. As a consequence, it's critical to be able to precisely and rapidly identify diabetes. People with type 2 diabetes often miss the first warning signs. For many years, they do not exhibit any symptoms. It is crucial to understand the possible health dangers of diabetes, and it may be quite helpful if the user can utilize a mobile app to do so. The Google Play Store and Apple App Store both include a wide variety of programs for managing diabetes, but many of them do not offer a mechanism to more precisely detect a user's health risk related to diabetes using machine learning techniques.

Research suggests that some individuals may be able to reverse it. Patients may be able to achieve and maintain normal blood sugar levels without taking medication by making dietary adjustments and losing weight. With medication and lifestyle changes, diabetic patients may see improvement in three to six months, according to diabetes experts.You can control your blood sugar levels by tailoring a diet plan using regional foods, modifying what and how much you consume, and monitoring your calorie needs.To better the body's composition Exercise and physical activity are additional crucial elements. They are crucial for muscle growth in addition to helping you burn more calories. The contraction of skeletal muscle stimulates glucose metabolism via an insulin-independent mechanism. The process of delivering glucose is made easier by an increase in blood flow to the active muscle areas. Regular aerobic exercise increases the production of glucose transporters.

RESEACRH GAP

We have mostly focused on similar research tasks within the same topic in the literature analysis. Our study's main goal is to design and develop smart, secure mobile applications for comprehensive diabetes lifestyle management. We began by doing some preliminary research on this using various secondary sources. We were able to learn about the things that were already there, which helped us to design our job plan.

Nabila Shahnaz Khan and Mehedi Hasan Muaz developed the mHealth application and provided a machine learning-based method for identifying whether a patient has diabetes or not using the Naive Bayes Classifier. Age, gender, BMI, and family history of diabetes were the four patient criteria they used to make the prediction. They conducted a survey to gather crucial data. They proposed a smartphone app called Diabetes Predictor, however it just has the capacity to predict diabetes, while DiaBeta has many more functions to help patients manage their diabetes-related lifestyles.

By absorbing light using NIR spectroscopy and projecting glucose to obtain the necessary penetration depth, researchers working under the direction of Mr. A.K. De Alwis have created a non-invasive method for monitoring blood glucose levels. They used a CJMCU OPT-11 photodiode to get blood glucose levels. Through the HC-05 Bluetooth module, glucose levels are detected, sent to the Arduino microcontroller, and connected to their DiabiTech Mobile app. Using fingerprint sensors, the creative team under the direction of Mr. Aveen Uthman Hassan has carried out non-invasive real-time monitoring of blood oxygen levels and heart rate for professionals. The suggested solution uses a Wi-Fi network to monitor heart rate and oxygen saturation using the Blynk app and NodeMCU.

METHODOLOGY

Overall research contains a mobile application, web-based system, and hardware device. Mobile application support both android and ios. Users can be diabetes and non-diabetes people. The system has three services which are diabetes prediction service, health risk identification service and meal and exercise recommendation service. These services are developed and deployed separately using microservice architecture. Mobile application has more user specific services like managing digital logbook, and web-based system contains services which are generic to the users (i.e., Diabetes Prediction, Health Risk Identification). The hardware device is connected with the mobile application, and it is used to measure Glucose, Heart Rate and SpO2 level.

“DieBeta” system has developed for 4 components as follows; Main objective of this research is to design and develop intelligent, secure smart mobile applications for complete diabetes lifestyle management. The Sub Objectives are as follows:

A. Diabetes Prediction
When predicting diabetes, the system used a data set which is the Pima Indian Diabetes dataset consisting of information on 768 patients (268 tested_positive instances and 500 tested_negative)[10]. The dataset consists of several medical predictive (independent) variables and an outcome that is one dependent variable. Independent variables are Glucose, Blood Pressure, Insulin, Skin Thickness, BMI, Age and Gender. For the prediabetes detection, the system used a data set that has been collected using direct questionnaires from the patients of Sylhet Diabetes Hospital in Sylhet, Bangladesh [9] and approved by a doctor. This dataset contains the sign and symptoms data of newly diabetic or would be diabetic patients. The dataset consists of 15 features and one target variable named class. In data analysis, it is essential to analyze the data and discover any features with missing values, as well as how they will affect the dataset. According to the analysis in the Pima Indian dataset there are a high percentage of missing values in key features for diabetes prediction. Such as insulin and skin thickness. Since these features are usually important predictors of presence of diabetes cannot remove the features, Instead, have removed the rows that contain the missing values for insulin. This will drastically reduce the dataset size, yet it will be increasing the model's accuracy. For the data standardization we have used Standard Scaler from sklearn. In Sylhet dataset, since the number of features are high, look at feature importance and select top 10 features for the model training. Since this is a classification problem we have used several classification algorithms (i.e. Logistic Regression, SVM Random Forest, XgBoost) and have selected the model with higher accuracy to predict diabetes. Then we deployed the model using flask and docker. For the container orchestration we used kubernetes. Users can access diabetes prediction services via both DiaBeta mobile application and DiaBeta web. They need to answer a questionnaire that includes necessary information for the diabetes prediction. After a questionnaire filled by the patient, the system predicts the result and sends it back to the user in a descriptive manner.

B. Recommend Diets and Exercise Plans
The ability of patients to achieve and maintain normal blood sugar levels may be improved by making dietary adjustments and losing weight. So, a healthy diet plan with a well-organized, balanced, and counted calorie intake is very important for controlling and preventing high blood glucose levels. When calculating, the system uses an API service to show the listed nutrition separately for the searched food items. The system displays results for a serving size of 100g by default for each item. Calories contained in the desired food item along with many other nutrients such as cholesterol, sodium, carbohydrate, fiber, and protein are shown in a table card. The total of every nutrition and calorie is displayed on the top as a separate table card. There is no one-size-fits-all approach to meal planning due to variations in custom, culture, preference, and financial situation. So, the system suggests several options for the recommendation of a diet plan, taking the necessary conditions of the DiaBeta users into consideration. The system asks for several factors when recommending the diet, such as users preference in veg or non-veg, type of cuisine as dessert, snack, etc. The system uses the feature extraction technique witch is TfidfVectorizer for recommendation. Then the system shows a sorted list of recommended diets. Diabetes is a significant, long-term condition in which the body cannot produce any or enough use the insulin that is produced effectively. Exercises can boost the number of insulin receptors in diabetes patients' external tissues, improve insulin sensitivity for improved function of the body's insulin, and cause skeletal muscles to burn glucose in tissues and blood to lower blood glucose levels. The system uses an API service to calculate estimated calories burned for searched exercises for a certain time period. For the recommendation of exercises system uses TfidfVectorizer for the feature extraction of the exercise dataset and according to user inputs system recommend exercises from the values retrieve from cosine similarity for the extract data.

C. Identify Health Risk & Give Health Education
In the Health Risk Identification part, a person must face a questionnaire before being identified for health risks. Uses data obtained by machine learning algorithms to determine if a user has a health risk. Since this indicates the health status of the person at risk for diabetes, it is possible to reduce the risk by giving the person a self-study about him or her. The system shows the user which organ of the body is at risk for the person at risk. The system predicts individual health risk levels, primarily targeting the eyes, heart, and kidneys. The data obtained in the first stage finds out whether the person has a health risk and the second shows the level of that risk. The system indicates that the at-risk person is seeking health counseling or, if the person is at high risk, they should see a physician related to the at-risk organ. This system differs from the others in that it automatically detects health risks and gives the best possible advice on how to avoid them. Therefore, all health tips and automation methods depend on the data provided by the patient. Users using the application can be informed about health education.

D. Measure Blood Oxygen Level, Heart Rate and Blood Glucose Level
The DiaBeta hardware device shown in figure 2 can be The DiaBeta hardware device shown in the picture above can be divided into two main parts a heart rate and blood oxygen level (SpO2) measuring system and a non-invasive blood glucose measuring system. The components used for the entire DiaBeta hardware device are MAX30100 sensor, NodeMCU, Jumper wires, Breadboard, OLED Display, two CJMCU OPT-101 photodiode, LED Bulbs, KY003 RGB SMD 5050 LED Module, and two 38 KHz 940nm IR Emitter Modules By placing the finger on the MAX30100 sensor, the heart rate and blood oxygen level are measured. NodeMCU is an ESP8266-based open-source software and hardware development environment that was cost-effective and includes RAM, wi-fi, CPU, and operating system. NodeMCU was used and the system was connected to the Firebase real-time database with the DiaBeta Mobile app via wi-fi. In addition, blood oxygen levels and heart rate measurements were also output through the OLED Display. NIR spectroscopy was chosen as the most appropriate optical source for non-invasive blood glucose measurement due to its maximum penetration depth in tissues. Many types of research using NIR emitters use NIR Indium Gallium Arsenide (InGaAs) sensors. The use of InGaAs sensors was high cost and high wavelengths around 1550 nm cause NIR radiation effects on the skin. Alternatively, in the research, two IR Emitter Modules with a wavelength of 940nm, which do not cause NIR radiation effects on the skin, and two CJMCU OPT-101 photodiodes, which are more economical and safer, were used to detect wavelengths in the region from 300 nm to 1100 nm. In addition, KY003 RGB SMD 5050 LED Module is different. It was capable of emitting 3 different light colors at a single wavelength in a single circuit. By placing the emitter on the top side of the finger and the photodiode on the bottom side of the finger, absorption of light is used to calculate the results.

RESEARCH OBJECTIVES

We have mostly focused on similar research tasks within the same topic in the literature analysis. Our study's main goal is to design and develop smart, secure mobile applications for comprehensive diabetes lifestyle management. We began by doing some preliminary research on this using various secondary sources. We were able to learn about the things that were already there, which helped us to design our job plan.

Nabila Shahnaz Khan and Mehedi Hasan Muaz developed the mHealth application and provided a machine learning-based method for identifying whether a patient has diabetes or not using the Naive Bayes Classifier. Age, gender, BMI, and family history of diabetes were the four patient criteria they used to make the prediction. They conducted a survey to gather crucial data. They proposed a smartphone app called Diabetes Predictor, however it just has the capacity to predict diabetes, while DiaBeta has many more functions to help patients manage their diabetes-related lifestyles.

By absorbing light using NIR spectroscopy and projecting glucose to obtain the necessary penetration depth, researchers working under the direction of Mr. A.K. De Alwis have created a non-invasive method for monitoring blood glucose levels. They used a CJMCU OPT-11 photodiode to get blood glucose levels. Through the HC-05 Bluetooth module, glucose levels are detected, sent to the Arduino microcontroller, and connected to their DiabiTech Mobile app. Using fingerprint sensors, the creative team under the direction of Mr. Aveen Uthman Hassan has carried out non-invasive real-time monitoring of blood oxygen levels and heart rate for professionals. The suggested solution uses a Wi-Fi network to monitor heart rate and oxygen saturation using the Blynk app and NodeMCU.

TOOL & TECHNOLOGIES

Machine learning tasks were implemented using python language on Jupiter Netbook. Developed the front-end using Flutter to implement all the tasks integrated with the databases and developed the machine learning algorithms integrated with the back end. Machine learning algorithms hosted on the Heroku platform. I utilize the Heroku tool since our backend response is in real-time, allowing for easy access to our service from anywhere at any time. The finest text editor for creating both web application and mobile apps is Visual Studio Code.

Python

Jupytor Notebook

Visual Studio Code

Anaconda

Heroku

Flutter

Dart

Firebase

Flask

Postman