**1. Introduction**

Obesity refers to excessive amounts of body fat. Obesity is not only caused by food genetics, the environment could also be a cause. The intake of energy and not consuming this energy through physical activity could also be a primary reason for obesity [1]. Obesity is the relationship between calorie intake and energy expenditure. It is a significant health issue associated with chronic illness and has a negative impact and long-term effects on patients and their families. As obesity is a risk factor for a number of diseases worldwide, it can be a threat to the world in the future. The Asia region is already dealing with malnutrition (as many cases have reported). Therefore, the number of obesity cases is increasing significantly with time [2].

**Citation:** Alsareii, S.A.; Shaf, A.; Ali, T.; Zafar, M.; Alamri, A.M.; AlAsmari, M.Y.; Irfan, M.; Awais, M. IoT Framework for a Decision-Making System of Obesity and Overweight Extrapolation among Children, Youths, and Adults. *Life* **2022**, *12*, 1414. https://doi.org/10.3390/ life12091414

Academic Editors: Tao Huang and Daniele Giansanti

Received: 8 August 2022 Accepted: 5 September 2022 Published: 10 September 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

Since 1975, the global obesity rate has increased thrice according to the World Health Organization (WHO) [3]. In 2013, the Indonesian Basic Health Research national survey (RISEKDAS) noted that obesity cases were rapidly increasing in Indonesia. Obesity can affect both men and women. The rate of obesity in adult men was 13.9%, 7.8%, and 19.7% in 2007, 2010, and 2013, respectively. In contrast, the rate of obesity in adult women was 14.8%, 15.5%, and 32.9% in 2007, 2010, and 2013, respectively [4]. However, in 2018, according to RISEKDAS (the same survey), the rates decreased to 14.5% in men and 29.3% in women [5].

The 2016 data show that the obesity rate has hit over 650 million people globally [6]. From age 18 and older, the ratio of people who are overweight increased to 39% [7]. Obesity and being overweight lead to other dangerous consequences that could lead to health anxiety. Obesity is the prime reason for significant lifestyle diseases, such as cancer, type II diabetes, lung disease, chronic pulmonary disease, and asthma.

Underdeveloped countries and populations are high victims of these diseases. NCDs (non-communicable diseases) and lifestyle diseases caused 36 million (63%) global deaths in 2008. Of these 36 million people, 80% were affected in underdeveloped countries and the middle class, 13% affected the upper class, and 29% affected those under age 60. Selected literature studies showed an annual increase of 10 million deaths due to NCDs. A survey from 2016 showed an increase of 71% (56.9 million), predicting 75% to 88.5% of deaths (until 2030) from NCDs in emerging countries, while the ratio predicted in developing countries is 65% [8]. Body mass index (BMI) is a primary risk element for the rise in diseases linked to sedentary lifestyles [9]. BMI helps in assessing body composition by calculating "weight/height".² However, BMI is considered a lousy sign of the proportion of body fat because BMI is dependent on age and does not count the fat on different body sites. According to the Institute of Medicine's 2012 report, there are population-based obesity prevention initiatives that address obesity and being overweight, such as a balanced diet, regular exercise, context- and setting-specific advice, and sound social norms [10].

There are several risk factors associated with obesity. In general, these factors are divided into categories, such as lifestyle factors (e.g., consuming junk food, alcohol, stress, and low physical activity), as well as demographic and socioeconomic elements (e.g., age, gender, marital status, place of residence, and genetic elements) [11]. Some risk factors can be avoided while others cannot. To implement an effective risk reduction strategy, the individual and population levels need to understand the factors that can be avoided [12]. The available data have helped numerous studies in exploring better approaches.

Epidemiological data modeling techniques (using machine learning) are popular in scholarly publications. These techniques can contribute to a better understanding of illness distribution, general health, risk identification, and health risk factors. There are several methods and algorithms available for this purpose [13]. The techniques require exact data classifications to assist in identifying risk detection from the information to lessen the danger signs and morbidity and mortality caused by obesity. Based on data showing compliance with dietary guidelines for obesity prevention, machine learning is applied to predict the likelihood of obesity [14]. Electronic health records are used in machine learning for predicting obesity in children, predicting obesogenic environments for children, and aggregating clinical data, such as metabolomic lipidomics and model drug dose responses [15].

In an online study conducted in Bangladesh (November 2020), 338 adults were examined. Sociodemographic statistics, health-related information, physical activity-related details, and nutrition measurements were all covered in the questionnaire. With two scenarios ('before' and 'during' the pandemic commencement) taken into consideration, inferential statistics (i.e., chi-square test, McNemar test) were employed to analyze the relationships between BMI and examined variables [16]. P0.05 was regarded as statistically significant. Results revealed that 30.5% of people were overweight "before" the COVID-19 pandemic and 34.9% of people were overweight "during" the pandemic. This suggests that 4.4% of the participants experienced significant weight gain after the pandemic started.

A recent report from Riyadh shows that 24.5% of women and 19% of men are suffering from obesity [17]. In 2021, the United Kingdom of Saudi Arabia showed significant increases in obesity rates. The ration varied in men and women but overall statistics showed that there were increases of 26.8%, 24%, 23.5%, 23.3%, 20.6%, 20.2%, 19.8%, 19.7%, and 14.2% in Riyadh, Makkah region, Hail (23.5 %), Al-Jouf, the northern border region, Madinah, Jazan, Tabuk, and Al-Baha, respectively, as depicted in Figure 1. This increase in people being overweight has led to an increase in diseases. The report further revealed that diseases such as obesity, diabetes, high BP, heart illness, stroke, and cancer are rising at ratios of (19.0%, 24.5%), (13.5%, 10.6%), (13.7%, 12.7%), (5.5%, 3.9%), (1.5%, 1.3%), and (1.3%, 1.8%) in both males and females, respectively, as shown in Figure 2.

**Figure 1.** Obesity rate in the provinces of the Saudi Kingdom.

**Figure 2.** Increased specific disease rate in males and females of the Saudi Kingdom.

Several machine learning algorithms are applied with several features to predict specific health conditions. A branch of machine learning known as ANN (artificial neural networks) correlates input parameters and corresponds to output data. ANN has reported several applications in engineering and medicine with variable success rates. Dugan et al. [18] employed artificial intelligence to predict childhood obesity. Six models were used in this research for the study. These models were naïve Bayes, random tree, ID3, j48, random forest, and Bayes net-trained. These models were applied to the clinical decision support system on CHICA. The results showed that ID3 performed well, giving a high ratio of accurate results at 85% and a sensitivity rate of 90%. Jindal et al. used

techniques for collective machine learning for obesity prediction. The prediction accuracy proposed for ensemble machine learning approaches was 89.68%. The generalized linear model, partial least squares, and random forest were used in the ensemble prediction through a Python interface.

Hammond et al. [15] used public records and electronic health records for the prediction of obesity in childhood. Several machine learning algorithms were trained for regression and binary classification. The results showed considerable accuracy in the first two years of data collection. The results showed that children at age five could become obese. To distinguish between low, medium, and high obesity, they used logistic regression (using a separate random forest classifier). They employed LASSO regression to 'prophesy' their continuous BMI values. The bootstrap was run 100 times to obtain a better performance of the model.

Obesity at the national level was predicted using data on food sales; see Dunstan et al. [19]. Three machine learning models were applied to the data obtained from seventy-nine countries. The authors researched basic information from the synergic nature of categories by analyzing food sales. They used five categories. The research considered 60% of countries for 10% (concerning the prevalence range). Moreover, 87% of countries projected the prevalence of obesity with an absolute error of less than 20%. The research showed that baked goods and flour were the most appropriate food categories for the prediction of obesity. Extreme gradient boosting, RF, and SVM were utilized for this model.

Singh and Tawfik [20] presented a machine learning model that might predict adolescent weight gain and obesity. In this study, seven machine learning methods were employed. J48 pruned tree, K-NN, bagging, and other algorithms were used, such as multi-layer perception and random forest. An unaltered and unbalanced dataset was used to vote on the effectiveness of all of the proposed algorithms. The MLP algorithm resulted in a 96% precision value. While the F1-score gave results of 93.96%. Gerl et al. [21] exhibited the use of large population cohorts for the prediction of different measures of obesity. A perplexing lipidomic signature was identified for BFP. A total of 73% of BFP variants were predicted based on age, gender, and lipidome, with the complete range of BFP having mistakes.

Montanezet al. [22] used publicly available genetic profiles and studied machine learning algorithms for predicting obesity. Many machine learning models were involved in this study, such as the SVM algorithm, decision tree, K-NN algorithm, and the decision rule for predicting susceptibility to chronic hepatitis with the help of SNP data. Of all the techniques, SVM produced the best results for the prediction model. According to the simulation findings, the SVM area was below the curve value of 90.5%.

Borrel and Samuel [23] worked on risk mortality and the US adult body mass index category. The effects of obesity and excess weight on the Cox proportional hazard regression were looked at to obtain the death prevalence. They calculated the rate of progress through time for all causes and the mortality rate dependent on peers at a normal weight. They also looked into the mortality rate of persons with obesity/were overweight and had cardiovascular disease. Their proposed results showed CVD caused death in obese adults (over 20% compared to normal-weight adults).

During the pandemic, the obesity rate increased due to lockdowns, and it become extremely important to have digital methods to monitor physical activities and the obeseness of people. Various challenges were involved in the understanding of risk factors and the obesity ratio. Traditionally, statistical analyses were used for understanding obesity, imposing independent linearity and a limited number of prediction sets. Therefore, this study focused on the different machine learning models for the risk identification of obesity. It evaluated the effectiveness of machine learning techniques, such as regression and classification on accessible data in order to compile a list of criteria that could be used to diagnose obesity and being overweight. These results helped us to design an IoT-enabled decision system that might be accessible worldwide where internet facilities are available. Thus, the paper provides the following contributions:


The remainder of the paper is organized as follows: the proposed machine learning algorithms and IoT system architecture are explained in Section 2. In Sections 3 and 4, the results and discussion are covered; the conclusion and future work are discussed in Section 5.
