See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/376517685

Diet Recommendation System Using Machine Learning

Preprint · December 2023

DOI: 10.13140/RG.2.2.20744.19207

CITATIONS READS

0 778

1 author:

Mohammed Shaheel
Bournemouth University
1 PUBLICATION 0 CITATIONS

SEE PROFILE

All content following this page was uploaded by Mohammed Shaheel on 15 December 2023.

The user has requested enhancement of the downloaded file.

 Diet Recommendation System Using Machine
Learning
Mohammed Shaheel
MSc:Digital Health and Artificial
Intelligence
Bournemouth University
Bournemouth,UK
kpmshaheel@hotmail.com

Abstract—Development of a recommendation system with help

of Machine learning techniques to promote healthy eating
habits and easy access for user for a tailored diet
recommendation which can help them with better food choice III. LITERATURE SURVEY
and improves health. An automized unsupervised model which Thi Ngoc Trang Tran et al. (2021) [1] demonstrate how
benefits users who don’t have access to a personal nutritionist their approach can be applied to provide an extensive
or may not afford such a one there by helping to improve overview of research on healthcare recommender systems:
general health. Furthermore, that it provides information for proposed
scenarios and methodologies. This includes advice on
I. INTRODUCTION nutrition, diets,medication plans, health services, and various
assistance from medical experts. To ensure that students fully
understand recommendation systems, they also provide real-
One of the leading problems around the world is unhealthy
world examples.
heating habits and lack of awareness of what to eat based on
their body conditions which further leads to development of Wenbin Yue et al [2] state that the purpose of his work is
chronic diseases. For instance, if you visit a food store, for to offer a thorough analysis of conventional recommendation
example, Subway, they just provide you with your allergen techniques and their uses in the healthcare industry. Three
list, which confuses the customer at times. Most of time main recommendation methodologies are stated here: content-
people are aware of their blood levels and sugar, but unaware based recommendation , collaborative filtering and hybrid
of content of food that can affect these levels. So here we try recommendation. After that, a discussion regarding five
an automized recommendation system with the help of application scenarios pertaining to health has been provided.
machine learning techniques to recommend various food These scenarios include, but are not limited to, dietary advice,
based on their contents with user input such as height, weight lifestyle advice, training advice, patient and physician
decision-making, and illness prediction. Ultimately, a number
etc and even to get a personalized diet plan based on this
of the most important issues facing this evolving field are
value. Many users from underdeveloped regions don’t have acknowledged, along with strong arguments.
access to personal nutritionists or may not be able to afford
such facility. Key advantage of such a system is that it can Gao et al. (2017) [3] introduced a framework for a
take consider of individual’s dietary restriction and personalized diet system. The system incorporated usage of
preferences and provide a tailor-made service for that Bayesian personalized ranking and matrix factorization
individual. For early training of the model, dataset named tounderstand what a specific user prefer over a large volume
food.com is available at Kaggle where it's possible to train of data. The results pointed out that performance of such a
and test the system model on thousands of food data values. system is much better than conventional collaborative filtering
Further, for interaction of User a web based front end will be techniques
developed for user input and recommendation output. The Nourishment Recommendation Framework (NRF),
which was proposed by Anonnya Banerjee et al. [4], gathers
II. PROBLEM STATEMENT user input from children, evaluates the information, and
generates an output that, in the end, produces a better diet plan.
Fast food consumption is alarmingly high, which has led to Its goal is to give kids between the ages of 8 and 13 nutrient-
the consumption of unhealthy food. This leads to high blood dense meals that are matched to their age, development,
pressure, diabetes, obesity, and other health problems. gender, and medical records.
Therefore, it is now absolutely necessary for people to have a
decently adjusted diet that is balanced, nutritious, and A machine learning technique is used by M. Geetha et al.
healthful. But not everyone in today's world has the time or [5] to analyze a person's body and predict their future health
status over the course of a calendar year based on their pre-
means to pay for a personal nutritionist or dietitian to assess
medical history. It provides diet recommendation systems that
their health and provide a balanced diet plan tailored to their
consider both historical and current food consumption data,
individual needs. We talked about the person's unhealthy and it makes more reliable diet report recommendations than
eating habits in this report and tried to create a it did previously.
recommendation system which automatically provides them
a way to improve their diets and live better lives. Ramni Harbir Singh, Sargam Maurya, et al.[6] used cosine
similarity and KNN to create a movie recommendation
system. This study describes a technique that offers users
generic recommendations based on a movie's genre and/or

XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE

popularity. Many deep learning strategies are used in the
Content-Based Recommender System's implementation. This
work also sheds light on the challenges faced by content-based
recommendation systems and our attempts to overcome them.

For diabetic patients, a Food Recommendation System

(FRS) [7] is suggested, which employs SelfOrganizing Map
and K-mean clustering for food clustering analysis. The
suggested method suggests the replacement foods based on
food parameters and nutrition.

IV. METHODOLOGY
The novelty in this approach is that the machine learning
model will learn based data on USDA food data base B. Data Preprocessing
where up-to-date, vendor-specific nutrition, allergen and
ingredient information of varous food items are • Normalization: Machine learning models perform
available. The idea was to run different machine learning better when all of the values are brought into a
algorithms on this data to learn what category of food single scale through the use of min-max
item would be appropriate for a particular user based on normalisation. Because the dataset contains a
values he given through the front end and predict a food variety of scale values—some are single digit
values, some features have two digest values, and
item suitable for the user.
some features have three digest values—
normalization is the process of condensing the
A. Datasets dataset to a single range of values. The min-max
scaling technique is used in this study to normalise
data that fall between 0 and 1 on the scale.
The recommendation system is based on USDA nutrition
information database. Every food item's nutrition information
is maintained in the USDA database. The USDA Foods • Data encoding is carried out following the removal
Database is available for state agencies and school districts to of inconsistent and duplicate values from the
access up-to-date, vendor-specific nutrition, allergen and dataset. Subsequently, the nominal attributes are
ingredient information for direct delivered USDA Foods for converted into numerical values. Performance will
child nutrition programs offered on the USDA FOOD be improved since machine learning models'
AVAILABLE LIST[8] .A sample of dataset being used are backend operations are carried out on numerical
shown in Fig 1 and Fig 2. values prior to being implemented using a machine
learning model. Before being given to data
encoding process, the non-numeric data included in
the database was converted to numeric data. Using
machine learning techniques, backend
computations were performed on numerical values
instead of nominal values prior to supplying input
to the suggested model..

C. Feature Selection.
For feature selection we make use of recursive feature
elimination approach. In this approach a ranking process is
introduced .Feature are ranked based on model’s coefficient
or feature’s importance approach further fits the model and
starts to eliminate each feature based on lowest ranking
through a recursive fashion. The process involve developing
a model with remaining features after repeatedly removing
least significant parts until desired number of features is
obtained.[9] RFE attempts to eliminate any dependencies or
collinearity that may exist in the model.
D.Machine Learning
1) K-means clustering is a popular unsupervised
machine learning algorithm in various applications, including
diet recommendation systems. K means clustering can be
used to put individuals who are similar together in a dataset
of users who have each disclosed information about their
dietary restrictions, allergies, and food preferences. For each
cluster, this can assist in creating customised meal plans,
recipes, and diet regimens. Each user is assigned to the closest
cluster using K-means clustering, which is based on how
similar they are.
2) Random Forest is a popular machine-learning
algorithm used in various content basedrecommendation
systems. It works especially well for tackling non-linear and
high-dimensional situations[10]. A supervised learning
algorithm that is a member of the decision tree family is
called Random Forest. To produce a more reliable and
accurate prediction model, Random Forest integrates several
decision trees, in contrast to conventional decision trees,
which only use one tree. Using Random Forest, a
recommendation system can build a model that forecasts a
user's preferences based on their prior behaviour as well as
the behaviour of users who are similar to them. Large datasets
can be analysed by the algorithm, which can then be utilised
to find trends and insights to provide tailored suggestions.
3) Decision Tree: A decision tree's primary objective is
to divide a population of data into more manageable
segments. They are supervised learning methods without
parameters that are applied to regression and classification.
The goal is to build a model that, by learning basic decision
rules deduced from the data's attributes, can forecast the value
of an output class.
4) The Naive Bayes approach, which is based on the
Bayes theorem, is an efficient supervised learning strategy for Fig. 3.Workflow of the proposed model
classification tasks. It’s a probabilistic classifier, which
means it generates predictions based on the chances of
something existing.The core idea of this classifier is based on
bayes theorem. In simple terms, a Naive Bayes classifier
assumes that the presence of a particular feature in a class is VI.MODEL EVALUATION
unrelated to the presence of any other feature. For model evaluation mainly four various assessment metrics
are used

Precision : It can be defined as the actual correct prediction

divided by the total number of predictions the model made.
To put it simply, our model predicts that 7 out of 10 patients
will have cancer, but only 3 of the predicted 7 patients
actually have heart disease, meaning that the precision in this
instance is 3/7 = 0.428.
V . WORKFLOW
Recall : Recall is determined by dividing the total number of
Following fig.3, show the proposed workflow of the true positives and false negatives by the number of true
recommendation system. Here dataset in form of csv obtained positives in a two-class classification problem. Let's say there
from USDA Database is employed to data preprocessing unit. are seven real cancer patients, but our model only predicts
Data preprocessing unit carries out three preprocessing stages that five of them will have cancer. In this scenario, the recall
: Normalization, encoding and visualization. Once is 5/7=0.714.
preprocessing is completed predictions are made by
recommendation system with help of all four machine Accuracy : It can be calculated by dividing the total number
learning algorithms. Further each model is evaluated basis on
of correctly classified examples by the total number of
various model assessment metrics
classified examples.

F1 Score : The precision and recall weighted average is called

the F1 score. It also takes into account false positives and
false negatives, as we are aware in precision and recall[11].
Generally speaking, F1 score is more valuable than accuracy,
particularly in cases where the class distribution is uneven.
When the cost of false positives and false negatives is built upon to get even a personalized diet plan for user.
comparable, accuracy performs best. It is preferable to Paper utilizes unsupervised learning techniques for training,
consider both Precision and Recall if the costs associated with testing and recording the response. At the end various
false positives and false negatives are significantly different. metrics will be used to record and test model’s accuracy
such as F1score, recall etc... Furthermore, develop a front-
User inputs his personal data such as age weight height, end, utilizing web development technologies for user
blood, sugar etc. This data is further run against all the four interaction with the system. The end goal is to design a
machine learning models which gives prediction. Based on system where users can get food and diet recommendations
model evaluation process which include above assessment based on input, such as height and weight through the
metrics carried out after predictions prediction of best model- interface.
based evaluation score is displayed for user.

X . ACKNOWLEDGMENT

VII. RESULT I would like to express my deepest gratitude to my Unit

Here diet recommendation system focuses on prediction leader for research methods and professional issues, Dr
developed by the four machine learning models incorporated Philip Davies, for his continuous guidance and expertise.
in the recommendation system.. The recommendation system His valuable insights and feedback have been instrumental
ask inputs from user end regarding user personal details. This in shaping the direction of this research. I would also like to
includes user’s Age,height,weight,blood,sugar level etc. This thank my colleagues for their dedicated efforts and
data is used to get a personalized food recommendation for the collaborative spirit.
particular user. Here the data given by user is run against all
the four machine learning models which gives prediction.
XI.REFERENCES
Based on model evaluation process carried out after
predictions, prediction of best model based evaluation score is [1] ]Thi Ngoc Trang Tran, Alexander Felfernig, Christoph Trattner and
Andreas Holzinger, “Recommender systems in the healthcare domain:
displayed for user state-of-the-art and research issues”, Journal of Intelligent Information
Systems, vol. 57, 2020, pp. 171–201, doi: 10.1007/s10844-020- 00633-
6.
VIII.PROPOSED FRONT END [2] W. Yue, Z. Wang, J. Zhang and X. Liu, "An Overview of
Recommendation Techniques and Their Applications in
An application can be further developed using the Fast API Healthcare", IEEE/CAA J. Autom. Sin., vol. 8, no. 4, pp. 701-717, 2021.
framework, which allows for the creation of fast and efficient H. Suhl, Eds. New York: Academic, 1963, pp. 271–350.
web APIs. When a user makes a request to the API (user data, [3] Jun Gao, Ninghao Liu, Mark Lawley, and Xia Hu, “An Interpretable
nutrition data...) the model is used to generate a list of Classification Framework for Information Extraction from Online
recommended food similar/suitable to his request (data) [4] A. Banerjee and N. Nigar, "Nourishment Recommendation Framework
which are then returned to the user via the API. for Children Using Machine Learning and Matching Algorithm", 2019
Int. Conf. Comput. Commun. Informatics ICCCI 2019, pp. 1-6, 2019
[5] M. Geetha, C. Saravanakumar, K. Ravikumar and V. Muthulakshmi,
The application's front-end is made with Streamlit. Streamlit "Human Body Analysis and Diet Recommendation System using
is an open source app framework in Python language[12]. It Machine Learning Techniques", 2021.
helps to create web apps for data science and machine [6] Ramni Harbir Singh, Sargam Maurya, Tanisha Tripathi, Tushar Narula,
learning in a short time. It is compatible with major Python Gaurav Srivastav,” Movie Recommendation System using Cosine
libraries such as scikit-learn, Keras, PyTorch, SymPy(latex), Similarity and KNN”,2020,((IJEAT), DOI:
NumPy, pandas, Matplotlib etc. For our case the front-end is 10.35940/ijeat.E9666.069520
composed of three web pages. The home page, which serves [7] Phanich, M., Pholkul, P., & Phimoltares, S., “Food recommendation
system using clustering analysis for diabetic patients,” in Proc. of
as an introduction and welcome page. The user can access the International Conference on Information Science and Applications, pp.
page with diet recommendations by using the left sidebar 1-8, IEEE, April 2010. Article
navigation. The user fills in his age, weight, height, blood [8] US Department Agriculture, https://www.fns.usda.gov/usda-fis/usda-
type, sugar level, and other details on the diet foods-database
recommendation page, and the page generates a diet [9] Avcontentteam,
recommendation based on the data provided. Additionally, https://www.analyticsvidhya.com/blog/2023/05/recursive-feature-
the user can further specify their food preferences using elimination/
nutritional values through the custom food recommendation.. [10] Tony yiu, https://towardsdatascience.com/understanding-random-
forest-58381e0602d2
IX .CONCLUSION [11] Dipesh Silwal, https://www.linkedin.com/pulse/confusion-matrix-
accuracy-precision-recall-f1-score-measures-silwal
[12] Zakaraia narjis, https://github.com/zakaria-narjis/Diet-
A content-based recommendation system which can produce Recommendation-System
food recommendations to a user based on his input. Further

View publication stats