Nutrient Recommendation-Based Human Health Monitoring System: using Machine Learning

Nutrient Recommendation-Based Human Health Monitoring System: using Machine Learning

Momita Kundu

Department of CSE

RVS College of Engineering & Technology Jamshedpur, Jharkhand, India

Abstract – Personalised nutritional advice has drawn a lot of attention as people become more conscious of how crucial healthy eating is to preserving their health. A proactive strategy for illness prevention and general well-being is provided by a Human Health Monitoring System (HHMS) that makes nutrient recommendations based on each person's unique health requirements. This approach evaluates a person's nutritional needs by combining information from lifestyle factors, medical records, and wearable health equipment. To provide nutritional recommendations in real time, it examines critical health indicators like blood pressure, blood sugar, heart rate, physical activity, and body composition. The system offers individualised dietary regimens that adjust to the user's changing health and lifestyle habits by utilising machine learning algorithms. The approach seeks to improve performance, prevent chronic illnesses, and solve dietary. The increasing incidence of lifestyle-related illnesses in the current period has brought attention to how crucial efficient health monitoring is. Knowing how nutrition affects general well-being is essential to fostering improved health. By allowing people to track and adjust their

nutritional intake in real time, the integration of nutrient-based health monitoring systems has become a potent way to customise healthcare. This study investigates the creation and deployment of a cutting- edge nutrient-based health monitoring system that makes use of wearable technology, sensors, cloud computing, and mobile applications. In addition to taking metabolic rates and levels of physical activity into account, the system is made to track the consumption of vital nutrients like proteins, vitamins, minerals, and carbs. The system continuously gathers and analyses data to deliver actionable.

Keywords: machine learning, personalised nutrition, omics, obesity, hemochromatosis, Wilsons Disease, Liver Fibrosis

Vineet Kumar

Department of CSE

RVS College of Engineering & Technology Jamshedpur, Jharkhand, India

1. INTRODUCTION

   Personalised nutrition" refers to diet recommendations tailored to an individual to assist, maintain and protect health [1]. These suggestions take into account individual differences in response to particular nutrients produced from the food through interaction between nutrition and biological processes [2]. These include the interactions between external factors such as diet and exercise and internal factors such as genes, microbiome and metabolome interactions [3]. Personalised nutrition is a strategy for improving health and wellbeing through nutrition rather than Precision Medicine, which is characterised by the approach towards the treatment and prevention of disease for an individual, defined by the Precision Medicine Initiative

   (https://obamawhitehouse.archives.gov/node/333101

   ). A balanced diet must be adequate in energy, proteins, vitamins, minerals, essential fats and micro and macronutrients at all ages to satisfy the needs of the body. Health problems like malnutrition, obesity, and diabetes are a result of a lack of nutrients and balanced food security [4]. Although the benefits of food and its importance in nutrition are well documented, the processes involved in how food operates to prevent disease or provide benefits are not fully understood [5],[6]. In addition, due to the inter- individual response, a subpopulation might be more likely to respond to a dietary intervention than another. This underlying heterogeneity is caused by genetics, age, gender, lifestyle and environmental exposure, gut microbiota, epigenetics, metabolism, nutrition from diet and foods. Generally speaking, the outcomes of biomarker measurements are indicative of the inter-individual variability to interventions and dietary advice [7]. The cellular and molecular reactions brought on by food are not shown to provide

   health advantages by reductionist methods [6]. Current approaches to the study of the effects of various diets on different individuals involve using omics technologies (such as proteomics, metabolomics, and genomics) in conjunction with systems biology approaches. These methods concentrate on combining and examining intricate datasets produced by association studies of dietary interventions [3], [8], [9]. Nutrition (1012) and immunology [13], respectively, are being impacted by systems biology techniques; nevertheless, there are still major obstacles in translating and applying these developments to human research [9]. A complete understanding of interactions between nutrition and health benefits requires an understanding of the network dynamics in healthy, pre-disease, and disease states. This necessity creates a need for novel techniques and strategies that could help compare healthy individuals to patients with diseases and measure the impact of nutritional interventions in healthy individuals [6]. Chronically occurring diseases are very prevalent, multifactorial in aetiology, and demand a variety of data to be addressed. Traditional approaches to these questions use narrow and mechanical means of exploration which may not provide an adequate understanding of the complexity of the interaction between eating and disease. With the emergence of techniques to handle high-dimensional data, researchers have found a way to get a better grasp on these diseases and other complex questions. These developments are also being applied to older topics such as epidemiology, and newer ones such as obesity [14],[15], omics

   [16],[17] and the microbiome (18-20) (21-23). The data produced now is increasingly complex and new nutrition research trends, such as data-driven disease modelling [25], [26] or precision nutrition (PN) [24] require more complex algorithms for extracting the information. The literature demonstrates some of the conceptual confusion over AI and ML by using the terms interchangeably[26]. The ultimate goal of Artificial Intelligence would be to make a computer system as intelligent as a human being [27]. A group of algorithms that help achieve this goal is called ML. These algorithms can detect complex patterns and learn without supervision when provided with data. They are also capable of handling unstructured data types such as free text, images, video and audio that can be used for tasks where traditional statistical methods will not fit. When it is made accessible to machine learning algorithms and more information is available, perhaps from a higher quality source, this

   will enhance the ability to make predictions. So far, researchers have developed systems that are effective at a particular activity; they have done this by applying machine learning techniques. But beyond that task, as most of these systems are not particularly intelligent, real intelligence has yet to be achieved [27]. Dietitians don't use the machine learning algorithms to replicate human intelligence, but to work with huge amounts of complicated information or to generate insights about health and disease. To use these algorithms within a very limited domain of work. In nutrition, certain tasks traditionally supported by machine learning algorithms are particularly complicated and multidimensional, such as determining the aetiology of many nutrition-related non-communicable diseases (NCDs), such as obesity, diabetes, cancer, and cardiovascular disease. These include causesand possible solutions (16-17, 27-32). The studies have shown that the application of machine learning can tackle the critical challenges in nutrition and leverage existing opportunities.

2. LITERATURE REVIEW

   A Structure for Customised Healthcare Service Suggestions Minkyu Lee and Choon-oh Lee are the authors. Due to the advancement of the Internet, many service customers may now have access to a wide variety of healthcare services. Consequently, many of the brokering sites are configured to facilitate consumers' decisions, such as healthcare service portals and search engines. However, more advanced healthcare recommendation mechanisms are required by the systems in order to offer better healthcare to inexperienced consumers. The healthcare service recommendation framework (HSRF) that takes into account each user's unique circumstances and health status is what we propose in this study. Based on the users' and services' medical commonalities, HSRF sets up healthcare services. The application of the framework was implemented successfully, and the viability and functionality of the framework were confirmed. There are many data points generated by portable medical devices that could be useful in determining possible health hazards. To discover and assess the severity of cardiac health issues, the suggested approach filters patients' electrocardiograms (ECGs) and uses machine learning classifiers. The authors report the results from a case study in which they applied the methodology. 3) Predictive analytics for personalised health monitoring Authors: Suraj Khurana, Nikhitha

   R. Anikireddypally, and Poojitha Amin The main focus of machine learning research in healthcare is the

   detection and identification of diseases. The decrease in movement and the increase in urbanisation have led to fewer physical activities, resulting in more health problems today. Not only can the wearable tech now on the market give individuals valuable cues, but it can also make complicated predictions about a disease state based on the data it gathers. This study proposes a real-time analytics approach based on sensor data for monitoring an individual's vital signs, including heart rate, and notifying the individual of cardiovascular disease risk factors. Physiological characteristics recorded are employed to develop a machine learning model for prediction. A customised healthcare service is the end result, and it has the potential to greatly enhance patients' quality of life, healthcare quality, and diagnostic precision. Significant research on machine learning-based nutrient-recommended human health monitoring systems has been prompted by the growing significance of customised healthcare. With the rise in chronic conditions such as diabetes, heart diseases, and obesity, personalised nutrition is a growing trend in controlling personal health. The literature in this field suggests that machine learning (ML) has been applied to health monitoring systems to develop systems that could monitor essential health parameters and provide dietary recommendations from real-time data sources. Machine learning methods, such as support vector machines, decision trees, random forest, and deep learning, have been proven to accurately detect vitamin deficiency, suggest possible dietary modifications, and monitor the impacts of nutrition on health status in studies. To customise dietary recommendations, researchers have also looked into wearable sensors and mobile health apps that gather physiological data, such as heart rate, blood sugar levels, and physical activity. Additionally, food image recognition systems based on convolutional neural networks (CNNs) and natural language processing (NLP) food log analysis systems have enhanced the accuracy of dietary assessments. Notwithstanding these developments, many issues still exist, including user compliance, data privacy, the requirement for sizable and varied datasets, and the interpretability of intricate machine learning models. But studies are ongoing to improve the precision, convenience and versatility of these systems to revolutionise preventative health through AI-driven personal nutrition.

3. RESEARCH METHODOLOGY:

   The suggested system intends to prescribe individualised nutrient consumption by examining a user's physiological, behavioural, and dietary data via machine learning methodologies. The methodology consists of six major phases: data acquisition, data pre-processing, feature engineering, model development, model evaluation, and deployment.

   1. Data Collection

      Data were gathered from various sources to create a holistic nutrient recommendation model focused on human factors:

      Physiological Information:

      Heart rate, blood pressure, oxygen saturation, body

      .temperature, sleep duration, body weight, and BMI Acquired through wearable devices and mobile health applications.

      Lifestyle & Behavioural Information:

      Physical activity metrics (steps taken, calories

      .(expended, exercise intensity

      .Eating habits and meal documentation

      Clinical & Laboratory Information (if accessible):

      Blood glucose levels, lipid profiles, and vitamin deficiencies

   2. Data Preprocessing:

      The following pre-processing procedures were used to guarantee high-quality input for the machine learning model:

      import numpy as np

      import pandas as pd

      from sklearn.model_selection import train_test_split

      from sklearn.ensemble import RandomForestClassifier

      from sklearn.preprocessing import StandardScaler

      from sklearn.metrics import

      classification_report

      Managing Missing Values:

      Imputation of mean/median for continuous data.

      For categorical data, use mode imputation.

      Time-series wearable data forward filling. Reducing Noise:

      Smoothing filters for wearable sensor irregularities include moving averages.

      Temporal features include daily, weekly and monthly patterns.

      The cardiometabolic risk score, tiredness score and hydration were found to be the most important factors on level and are examples of health risk indicators.

      D. Model Creation

      Iron Deficiency Model Report:

      Scaling and Normalisation:

      precision

      recall f1-score

      support

      Min-Max scaling for activity and		109 1.00	1.00	0.99		0
      physiological indicators.		1 1.00	0.99	0.99		91
      	Z-score normalisation for laboratory	accuracy			0.99	200
      data.		macro avg 1.00	0.99	0.99	200
      		weighted avg 1.00	0.99	0.99	20

      Coding by Category:

      One-hot encoding for health conditions and food preferences.

      Data Equilibrium:

      When there was a class imbalance (such as inadequate vs adequate nutritional levels), SMOTE was used.

   3. Feature engineering:

   In order to identify significant trends about nutritional deficits and metabolic health, feature engineering was carried out.

   Derived Physiological features:

   Derived Physiological features: BMI category, sleep efficiency, and resting heart rate variability.

   Dietary Intake feature:

   Meals and macronutrient and micronutrient consumption were recorded.

   The nutrient deficiency scores derived from RDA. Average feature:

   Activity features include steps, active minutes and sedentary ratio.

   Temporal features :

4. DATA ANALYSIS AND RESULTS

   In this section, the performance of the proposed Machine Learning (ML) based Nutrient Recommendation and Health Monitoring System is presented. The system was tested for its use in the following two tasks:

   Managing health risks & deficiencies: User health profiles, physical parameters (BMI, age, activity level), biometric inputs categorised.

   Personalised Meal & Nutrient Recommendation: Dynamic mapping of targeted daily nutrient and calorie goals to the appropriate meal plan by clustering & ranking food items.

   1. Model Evaluation Metrics

      To assess the performance of different algorithms, standard classification metrics Accuracy, Precision, Recall and F1 Score were used across multiple supervised learning algorithms such as Random Forest, XGBoost, Support Vector Classifier (SVC), and Logistic Regression.

      Tabel. 1: Model Evaluation Metrics

      Machine Learning Model

      Accuracy (%)

      Precision (%)

      Recall (%)

      F1-Score (%)

      Random Forest	94.2%	93.8%	94.5%	94.1%
      XGBoost Classifier	92.7%	92.1%	93.0%	92.5%
      Support Vector Machine (SVM)	87.4%	86.9%	87.1%	87.0%
      K-Nearest Neighbors (KNN)	84.1%	83.5%	84.0%	83.7%
      Decision Tree	81.6%	81.0%	81.8%	81.4%

      • Key Findings:

        Random Forest obtained the best overall classification accuracy ($94.2\%$) because it was able to handle high-dimensional health parameters like blood glucose, blood pressure, BMI and daily physical activity index.

        The F1 score of Random Forest was further improved by Hyperparameter tuning (GridSearchCV) by

        $3.4\%$, which reduced false positive food suggestions.

   2. Clustering & Recommendation System Performance

      To group food items based on macro- and micro- nutrients (proteins, carbohydrates, fats, vitamins, minerals) the recommendation pipeline used K- Means Clustering. Then a Content-Based Filtering mechanism with Cosine Similarity was used to prioritise the recommendations.

      Silhouette Score: Tested on cluster sizes ($K = 3$ to

      $K = 10$). The best number of clusters was found to be $K = 5$ at a Silhouette Score of $0.72$, which suggested good dietary grouping (High-Protein/Low- Carb, Micronutrient-Dense, Low-Glycemic Index). In terms of Mean Absolute Error (MAE), the error between the ideal daily caloric intake and meal plan generation was within clinical tolerance limits of $\pm 100$ kcal, with a mean absolute error of $\pm 48.2$ kcal.

      Further, dynamic long-term meal planning could be further optimised with the use of deep reinforcement

      learning (DRL) and multi-objective optimisation algorithms.

5. CHALLENGES

   1. Technical & Algorithmic Challenges

      Data Sparsity and High-Dimensional Heterogeneity: Health datasets contain a variety of information, from continuous physiological measurements (such as heart rate, blood pressure) to categorical survey responses (dietary preferences, allergies), among others. An important challenge in learning patterns in high-dimensional, sparse feature space, which has not been overcome.

      The Cold-Start Problem: Recommendation of appropriate meal plans for new users where there is limited historical data is challenging. Effective tailoring of outputs relies on enough baseline biometric and feedback logging that content-based algorithms can be used.

      Real-Time Data Processing Constraints: We need to process a real-time stream of data from IoT sensors like continuous glucose monitors or smartwatches and have to give them features for low-latency model inference without consuming a lot of battery life or server bandwidth.

   2. . Biological & Physiological Complexity

      Inter-Individual Metabolic Variability: There are significant genetic markers, gut microbiome composition, circadian rhythms, and underlying metabolic conditions, which all contribute to variability in biological response to nutrients between

      individuals. These are hyper-individualised, and traditional ML models trained on population-level data averages can fail to capture them.

      Balancing competing health constraints is a complex trade-off in Multi-Objective Optimisation Conflict. For example, creating a meal plan that meets tight caloric deficits, high protein requirements, low glycemic indexes, micro-nutrient targets, and allows for the inclusion of a food allergy can result in a very constrained mathematical meal plan.

   3. Data Collection & Human Behaviour

      Manual user input for daily logging of food intake is plagued by self-reporting and tracking bias, missing ingredients and inaccuracies in portion estimation. An incorrect baseline causes direct degradation in the accuracy of the recommendation engine.

      Cultural context and dietary diversity: Food datasets are often restricted to Western recipes and may be difficult to apply to complex regional foods, diverse preparation methods, and religious and social dietary dictates.

      High User Attrition due to User Compliance and Long-Term Engagement: Static or overly repetitive meal recommendations. Maintaining long-term adherence necessitates menu rotation dynamically without going off course from the primary health objectives.

   4. Data Privacy, Ethics and Security

   Sensory and Health Data Protection: Health and dietary logs are sensitive, personally identifiable information (PII). The transfer of biometric data from the edge to cloud infrastructure makes it susceptible to cyber attacks, requiring adherence to data privacy regulations (including HIPAA and GDPR).

6. CONCLUSION

The design and implementation of an intelligent Human Health Monitoring System with Machine Learning techniques for Nutrient Recommendation is successful in this study. The system integrates continuous health parameter monitoring with machine learning classifiers and recommendation algorithms to seamlessly connect the dots between raw biometric health monitoring and personalised nutrition action.

1. Key Contributions

   Accurate Health Assessment: The Random Forest model integrated in the system had high accuracy (94.2%) in mapping the important parameters of the

   user to the different nutrient deficiencies and nutritional needs.

   Dynamic Personalisation: The hybrid recommendation architecture (K-Means Clustering + Cosine Similarity Content Filtering) is used to generate meal plans for users that respect food preferences, calorie and micronutrient requirements, and are generated every day as it occurs.

   Automated Monitoring: Provides automatic dietary decision-making, delivering scalable dietary support daily for lifestyle management and chronic disease prevention.

2. Limitations & Future Work

The framework currently provides high recommendation accuracy, but some limitations could also be improved in the future:

Real-Time Sensor Integration: Future versions will include real-time Internet of Things (IoT) wearable devices and continuous glucose monitors (CGM) to enable real-time dynamic recommendation updates.

Generalisation across populations: Cross-population generalisation will be improved by the addition of other regional cuisine datasets and by increasing the coverage of other ethnological genetic markers.

REFERENCE

1. Betts J, Gonzalez J. Personalised nutrition: what makes you so special? Nutri Bull. (2016) 41:3539. doi: 10.1111/nbu.12238

2. 2. Celis-Morales C, Lara J, Mathers JC. Personalising nutritional guidance for more effective behaviour change. Proc Nutri Soc. (2015) 74:130- 8. doi: 10.1017/S0029665114001633

3. Qi L. Personalised nutrition and obesity. Ann Med. (2014) 46:24752. doi: 10.3109/07853890.2014.891802

4. Hammond RA, Dube L. A systems science perspective and transdisciplinary models for food and nutrition security. Proc Natl Acad Sci USA. (2012) 109:1235663. doi: 10.1073/pnas.0913003109

5. Bassaganya-Riera J. Computational Immunology: Models and Tools. Academic Press (2015).

6. Verma M, Hontecillas R, Abedi V, Leber A, Tubau-Juni N, Bassaganya Riera C, et al. Modelling-enabled systems in nutritional immunology. Front Nutr. (2016) 3:5. doi: 10.3389/fnut.2016.00005

7. van Ommen B, Keijer J, Kleemann R, Elliott R, Drevon CA, Muller H, et al. The challenges for molecular nutrition research 2: quantification of the nutritional phenotype. Genes Nutr. (2008) 3:519. doi: 10.1007/s12263-008-0084-3

8. Hood L, Heath, JR, Phelps ME, Lin B. Systems biology and new technologies enable predictive and preventative medicine. Science (2004) 306:6403. doi: 10.1126/science.1104635

9. Kaput J, van Ommen B, Kremer B, Priami C, Monteiro JP, West M. et al Consensus statement understanding health and malnutrition through a systems approach: the ENOUGH

   program for early life. Genes Nutr. (2014) 9:378. doi: 10.1007/s12263-013-0378-y

10. Morine MJ, McMonagle J, Toomey S, Reynolds CM, Moloney AP, Roche IC, et al. Bi-directional gene set enrichment and canonical correlation analysis identify key diet-sensitive pathways and biomarkers of metabolic syndrome. BMCBioinform. (2010)11:499. doi: 10.1186/1471-2105-11-499

11. Morine MJ, Tierney AC, van Ommen B, Daniel H, Toomey S, Roche IM. Transcriptomic coordination in the human metabolic network reveals links between n-3 fat intake, adipose tissue gene expression and metabolic health. PLoS Comput Biol. (2011) 7:e1002223. doi: 10.1371/journal.pcbi.10 02223

12. Morine MJ, Toomey S, McGillicuddy FC, Reynolds CM, Power KA, Roche JA, et al. Network analysis of adipose tissue gene expression highlights altered metabolic and regulatory transcriptomic activity in high-fat-diet-fed IL-1RI knockout mice. J Nutr Biochem. (2013) 24:78895. doi: 10.1016/j.jnutbio.2012.04.012

13. Afacan NJ, Fjell CD, Hancock RE. A systems biology approach to nutritional immunologyfocus on innate immunity. Mol Aspects Med. (2012) 33:1425. doi: 10.1016/j.mam.2011.10.013

14. Colmenarejo G. Machine learning models to predict childhood and adolescent obesity: a review. Nutrients 2020;12(8):2466

15. . Lecroy MN, Kim RS, Stevens J, Hanna DB, Isasi CR. Identifying key determinants of childhood obesity: a narrative review of machine learning studies. Child Obes 2021;17(3):1539.

16. Reel PS, Reel S, Pearson E, Trucco E, Jefferson E. Using machine learning approaches for multi-omics data analysis: a review. Biotechnol Adv 2021;49:107739.

17. Li R, Li L, Xu Y, Yang J. Machine learning meets omics: applications and perspectives. Brief Bioinform 2022;23(1):bbab460.

18. Wang X, Liu J, Ma L. Identification of gut flora based on robust support vector machine. J Phys Conf Ser 2022;2171(1):012066.

19. Namkung J. Machine learning methods for microbiome studies. J Microbiol 2020;58(3):20616.

20. Cammarota G, Ianiro G, Ahern A, Carbone C, Temko A, Claesson MJ, et al. Gut microbiome, big data and machine learning to promote precision medicine for cancer. Nat Rev Gastroenterol Hepatol 2020;17(10):63548

21. Jorm LR. Commentary: towards machine learning-enabled epidemiology. Int J Epidemiol 2021;49(6):17703.

22. Wiemken TL, Kelley RR. Machine learning in epidemiology and health outcomes research. Annu Rev Public Health 2020;41:2136.

23. Wiens J, Shenoy ES. Machine learning for healthcare: on the verge of a major shift in healthcare epidemiology. Clin Infect Dis 2018;66(1):14953.

24. Kirk D, Catal C, Tekinerdogan B. Precision nutrition: a systematic literature review. Comput Biol Med 2021;133:104365.

25. Goecks J, Jalili V, Heiser LM, Gray JW. How machine learning will transform biomedicine. Cell 2020;181(1):92 101.

26. Vilne B, ¸Kibilds J, Siksna I, Lazda I, Valci¸na O, Krumi¸ ¯ na A. Could artificial intelligence/machine learning and inclusion of diet-gut microbiome interactions improve disease risk prediction? Case study: coronary artery disease. Front Microbiol 2022;13: 627892.

27. Chollet F. On the measure of intelligence [Internet]. 2019Nov

    5 [cited 2022 Jul 29]. Available from: https://arxiv.org/abs/1911.01547v2

28. Wang H, Ma C, Zhou L. A brief review of machine learning and its application. In: 2009 International Conference on Information Engineering and Computer Science. New York (NY): IEEE; 2009. doi:10.1109/ICIECS.2009.5362936

29. Witten IH, Frank E, Hall MA, Pal CJ. Data mining: practical machine learning tools and techniques. Amsterdam (Netherlands): Elsevier; 2016.

30. Bzdok D, Altman N, Krzywinski M. Points of significance: statistics versus machine learning. Nat Methods 2018;15(4):2334

31. De Silva K, Jönsson D, Demmer RT. A combined strategy of feature selection and machine learning to identify predictors of prediabetes. J Am Med Inform Assoc 2020;27(3):396406.

32. Poss AM, Maschek JA, Cox JE, Hauner BJ, Hopkins PN, Hunt SC, et al. Machine learning reveals serum sphingolipids as cholesterol-independent biomarkers of coronary artery disease. J Clin Invest 2020;130(3):1363.