Development Edge Device Monitoring System Stunting and Malnutrition in Golden age 0 – 5 years Integrated with AI

: Golden age is the best period for a child's growth, monitoring of growth and development must be carried out regularly. Growth and development disorders in children are stunting and malnutrition. This incident also prompted the government through the Ministry of Health to create a special program towards a golden Indonesia 2045 to monitor stunting and disease, especially in children. The prevalence of stunting decreased to 21.6% from 24.4% in 2022. Early detection of stunting and malnutrition, where the research object is children aged 0-5 years. This prototype was built using a load cell sensor, a study of the use of optical sensors and ultrasonic sensors to measure body height, and a MAX sensor to detect children's anemia. The integration of this tool combines IoT and AI. The results obtained to validate the use of load cells have a reading error of 0.01% with an accuracy of 99%. Comparison using optical sensors and ultrasonic sensors. Optical sensors have result average error of 0.01, accuracy 98.99%, ultrasonic sensors error was 0. 15 with 85% accuracy. To measure malnutrition, the anemia parameters were processed using the Dense Neural Network (DNN) model with 256 neurons showing an accuracy of 98.03%.


Introduction
To support the government's program towards the golden generation of 2045, one of them is to overcome the problems of malnutrition and stunting.Health indicators in society are monitoring nutrition, immunization and child growth (kemenkes RI, 2010).Imbalance of nutrition in the body triggers stunting, and results in disruption of body functions and becomes very dangerous if this is not monitored properly.Parental knowledge, environment and economic level really support the fulfillment of nutrition during the golden period of children's growth and development in the first 1000 days.Children need adequate nutritional intake to optimize their growth.
Data from the World Bank states that the workforce that experienced stunting in their infancy reached 54%.This means that 54% of the current workforce are stunting survivors.This is what makes stunting a serious concern for the government (BKKBN, 2022).And it is reported that in 2022, the prevalence of stunting in Indonesia will begin to decline to 21.6% from 24.4% (kemenkes RI, 2023), shown in Figure 1, the distribution of stunting in Indonesia (Fundrika, 2022).
Sustainable Development Goals (SDGs) in the health sector state that there will be no more incidents of malnutrition, in accordance with the 2025 international target of reducing the incidence of stunting and wasting, providing good nutritional balance for pregnant and breastfeeding women, children and teenagers (Unicef, 2022).Stunting is short height, it can even be very short if seen based on height compared to age (H/A) with a zscore value between -3 SD to <-2 SD in children under 5 years (WHO, 2020)(humas bkpk, n.d.).There are 4 ways to solve the problem in this case, namely (P2PTM Kemenkes RI, 2018): 1. Provide exclusive breastfeeding to babies up to 6 months old; 2. Monitor children's development and take them to posyandu regularly; 3. Regularly consume Blood Supplement Tablets (BST); 4. Provide MPASI which is nutritious and rich in animal protein for babies over 6 months old.
Figure 1.Stunting Distribution (Fundrika, 2022) Nutritional status (H/A) and adequate levels of energy, protein and iron are significantly related to children's gross motor development (Anggorowati, 2013), malnutrition is closely related to child development.Nutritional status (H/A) and adequate levels of energy, protein and iron are significantly related to children's gross motor development (Clark,208 C.E.).From this background, an innovative tool was created to record height and weight as a parameter for stunting and hemoglobin levels to prevent animea as a parameter for malnutrition in children which were integrated into one edge device.
Previous research has been carried out to detect stunting and anemia, but combining both parameters and presenting them in one interactive application is not yet available.Several methods have been used to detect animea, one of which is the POCT method (Nidianti et al., 2019), then there is also testing using premier data which is carried out at posyandu (Nurlita et al., 2021) as well as several studies using stunting detection methods with AI (Roni et al., 2022), IoT (Rafaleony Berlian Putri Widodo, 2022), for animea detection (Ningsih et al., 2019)(REZZA, nd), development of a stunting detection application (Zenit Rohmaningrum, 2021), so far the algorithm still uses machine learning for animea detection using the SVM, ANN, DT, KNN method (Bitew et al., 2022) Technology built using edge devices by integrating IoT (Internet of Things) and Dense Neural Network (DNN) to process animea data into a model that is ready to use.The sensors used include MAX sensors, load cells, optical sensors.Equipped with a web-based application that can be accessed anytime and anywhere without having to install an application along with system reminders.And parents can monitor children's nutritional levels, indications of stunting and malnutrition in one system.

Method System Hardware Design
To research indications of stunting and malnutrition in children aged 0-5 years, this was carried out at the posyandu in collaboration with PKK RW 12 Karanggan Bogor mothers.To build this system there are several steps, namely hardware design and user interface design.
The hardware design for the stunting detection system was made using a load cell sensor to measure body weight, comparing the performance of the VL53LO sensor with ultrasonic to get the best value, and also the MAX 30100 sensor to detect animea levels, in Figure 2 shows the system created, Figure 3

User Interface Design
The data taken by the sensors will be sent directly to the cloud server where the data will then be processed and displayed on web.

Dataset Creation
The stages of creating the machine learning model that will be used in this research, where the machine learning model uses the Python programming language with the DNN model, is targeted for the model to reach R2 > 0.92.Table 1 provides the formation of the animea dataset.

Standardization of measurements
To determine whether a child is stunted and has good nutrition or not (Ministry of Health, 2020) shown in Table 2.

Analysis
System evaluation is divided into 2, namely model evaluation for animea data processing and also sensor evaluation for weight and height reduction.For evaluation of weight and height measurements, data validation tests will be used with calibrated measuring instruments and the two values are compared to obtain error values and accuracy values. (1) To evaluate AI using several methods.Evaluation of the model's algorithm performance in predicting is carried out using a loss function.The loss function determines the error distance between the algorithm's predicted results and the expected output.The method used is: 1. Mean Absolute Error (MAE) MAE or Mean Absolute Error shows the average error value which is the error between the actual value and the predicted value.MAE itself is generally used to measure prediction error in time series analysis.The formula for calculating MAE is: Y' is the predicted value, Y is the true value and n is the amount of data.

Mean Squared Error (MSE)
MSE measures the average value of the total squared difference between the model predicted value and the actual value(Bayangkari Karno, 2020) The formula for calculating MSE is:

Results and Discussion
Testing is carried out to test sensor readings and also validate whether the readings are good enough or not, where the sensor data tested is max sensor data, load cell sensor and VL53LO optical and ultrasonic sensors.Important notes in storing tools and using tools must be in the range 0-40 with humidity 10-90% RH.℃.

Load Cell sensor test results
This test was carried out using a weighing pendulum which varied from 500 grams to 2kg with the results obtained as in Table 3. Sensor testing was carried out at a temperature of 36 ℃.From the test results on the sensor with 10 different data with a sampling interval every 1 minute, the average error result was 0.01 and the average measurement accuracy was 99%.From this data, it can be ensured that the load cell sensor readings are running well and had high accuracy.

Results of Testing the Effectiveness of the VL35LO Sensor with an Ultrasonic Sensor
This test aims to compare the optimization of Height measurements, as shown in Tables 4 and 5 respectively.From the test results on both sensors with varying distance ranges from 10 cm to 100 cm, the results were obtained that using the VL53LO sensor had better reading and was still very optimal up to a distance of 100 cm, but for the Ultrasonic sensor, even though we bought it with sensor specifications, it could measure up to 2 meters.The furthest distance was proven in measurements when the distance was 75 cm.The reading started to have an error of 0.31 and the sensor accuracy was only 65%.This is not good for the tool being built because this prototype requires a high level of accuracy, so the VL53LO sensor is considered good enough to support height measurements with high sensor accuracy of 98.99%.Testing training data results for the max30100 sensor This test was carried out using coding directly on the system and obtained results as in Table 6.From the results of training, it can be seen that in MSE with 256 neurons the results shown are quite optimal and sufficient to be a model of the animea system, so that the model will use 256 neurons.

Testing Integrated System
After the entire system is installed and integrated with each other as in Figure 5, we tested the tool.Figure 6 shows the results of measuring stunting and nutritional status of children.Table 7 provide data for the reading results and precision of animea levels in children.From the test results, an error result of 0.02 was obtained and the reading accuracy was 98.03%.

Conclusion
After carrying out all the stages, it was found that for testing the load cell sensor which is intended for measuring children's weight, an average sensor accuracy of 99% was obtained.For measuring body height, from the experimental results it was found that the use of the VL53L0 sensor was more optimal with an average sensor accuracy of 98%.When the system integrate has result for accuracy is 99%.quite good results were obtained where the results could be read on the system and could be displayed well and for the results of reading animea levels using DNN with 256 neurons, the results obtained an average accuracy in reading of 98.03%.From the users and also the companions who took part in the testing, it was said that the tool was easy to use because one direct measurement can directly measure 3 parameters and shorten the measurement time, especially if the child has a feeling of trauma or fear when taking measurements.
is Realization of hardware for children aged 0-1 year, and for Figure4is hardware for children aged 2-5 years.

Figure 2 .
Figure 2. hardware design system

Table 2 .
Standardization of Measurements

Table 3 .
Weight Measurement Results

Table 4 .
Testing with the VL350 optical sensor.

Table 6 .
Training data results

Table 7 .
Animea sensor reading with Max 30100 sensor