Diet Recommendation using Predictive Learning

Approaches
2022 3rd International Conference on Issues and Challenges in Intelligent Computing Techniques (ICICT) | 978-1-6654-8268-4/22/$31.00 ©2022 IEEE | DOI: 10.1109/ICICT55121.2022.10064538

Anjali Jain Alka Singhal

Department of CSE, Department of CSE,
Jaypee Institute of Information Technology, Jaypee Institute of Information Technology,
Noida, India Noida, India
KIET Group of Institutions, Delhi-NCR Ghaziabad [email protected]
[email protected]

Abstract— In the modern world, people's health is one of Diet advice is in high demand right now since so many
their top priorities. As a result of inadequate nutrition, many people are asking for it and so much research is being done
people in the modern world are afflicted with a variety of
on the topic.
diseases. It is urgently necessary to make timely and easy
recommendations for a balanced diet, which is quite A person's diet depends on a number of variables, including
challenging. Time is one of the major restraints in today's
world, therefore authors created a system for recommending nutrition, accessibility, disease, and technology. A person's
diets to people based on their unique health needs, whether diet has an impact on how their body morphologically
they are trying to gain weight, lose weight, or get a health changes. To lose weight, people used to fast, skip meals,
checkup. The system is built on machine learning classifiers and practise intermittent fasting. Therefore, it is necessary
including Random Forest, Support Vector Machine, Ada to offer a suitable and balanced food plan so that one can
Boost, and Gradient, as well as clustering methods like K-
means. Boost. After calculating Body Mass Index and the adhere to it. As the need for and consumption of The advice
user's desire for diet based on their health state (overweight, of a diet should be made with these considerations in mind
underweight, or healthy), the suggested system makes meal because the optimal diet for each individual varies
recommendations based on the user's age, height, and weight. depending on their age, weight, height, disease, and gender,
The proposed paper summarises numerous similar works and among other things. The author of this research will
evaluates the effectiveness of the suggested strategy. The
performance comparison is displayed in terms of recall, recall prescribe diets using a machine learning algorithm. The
accuracy, precision, and f1-measure. need for dietary advice has expanded to the point where
numerous diet regimens, including the low-carb, low-fat,
Keywords— DRS, Predictive Learning, BMI. keto, paleo, and vegan diets, have been developed. Every
diet has its own advantages, such as helping people lose
weight, gain weight, or stay in shape. Numerous diseases
I. INTRODUCTION that used to be caused by social hierarchy are now caused
People in the modern world are too busy to take care of their by inadequate nutrition and biological dysfunction as a
health or eat healthfully. People need someone to give them result of the tremendous changes in the environment,
a healthy diet plan because they frequently skip meals at climate, and way of life of humans. Most people desire to
specific times. Numerous elements, including nutrition, eat better and be healthier, which takes a lot of work and
heredity, and sleep, have an impact on a person's health. planning of one's schedule [1][9].
People used to follow diet advice from many online sites
Organized and efficient machine learning algorithms
because the internet is so overflowing with information. A are utilised to prescribe the right amount of nutrients,
balanced diet that promotes activity and a long life is the calories, fat, and carbohydrates for a diet that will keep you
most important aspect of human existence. A balanced diet fit and healthy. Numerous other techniques have been the
is one that contains the right ratio of carbs, lipids, proteins, subject of research, such as similar user behaviour, user
vitamins, minerals, and sugar, among other things. Diet preferences for item features, hybrid filtering (HF),
advice is in high demand right now since so many people knowledge-based filtering (KBF), ant colony algorithm,
are asking for it and so much research is being done on the classification, clustering, decision tree, logistic regression,
natural language processing (NLP), inductive logic
topic. A person's diet depends on a number of variables, programming (ILP), ontologies, sparse canonical
including nutrition, accessibility, disease, and technology. correlation, support vector machines, semantic
A person's diet has an impact on how their body technologies, multi-criteria decision making, and graph-
morphologically changes. To lose weight, people used to based recommendation[2]. These are the fundamental
fast, skip meals, and practise intermittent fasting. recommendation strategies used in the different areas of
Therefore, it is necessary to offer a suitable and balanced healthcare. The advantages and disadvantages of various
food plan so that one can adhere to it. There is a lot of filtering methods have also been covered by authors. To
forecast the optimal meal menu for a given health status, the
inquiry and research being done for diet recommendations.
author of the proposed study used machine learning
1

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SILCHAR. Downloaded on August 30,2024 at 06:46:16 UTC from IEEE Xplore. Restrictions apply.
 classifiers such Random Forest, Support Vector Machine, youngsters between the ages of 3 and 6 years old, a different
AdaBoost, and gradient boost [3]. authors created a sensor-based smart plate that measures the
weight of food both before and after meals.

II. RELATED WORK The suggested approaches are intended to track children's
growth indices and dietary deficits. The results and analysis
Having a diet that is well-balanced in nutrients is the key to
of the necessary amount of nutrition led to the
living a healthy life. Because eating healthfully is a need
recommendation of the nutrient-dense menus. There are
today. People used to diet to maintain their health, gain
numerous other methods that different authors employ
weight, or lose weight. However, these kinds of diets,
when recommending a cuisine or diet.
procedures, and restrictions are bad for their physical and
mental health. A person should consume in accordance with Jiang et al. (2020) employed a graph mining method based
their desires, nutritional requirements, and ingredient on random walks that can also be applied to address user
needs. There are many recommendation strategies that have health demands. The Food-Nutrition-Recipe Graph
been applied to diet recommendations, such as similar user (FNRG), which was developed in this study and proved
behavior, which suggests products based on the interests helpful for those with chronic diabetes, is a heterogeneous
and preferences of other users who are similar to the one in graph that integrates data from Food Data Central (FDC),
question. Similar user behavior and preferences for item recipe websites, and scientific literature. (Chena et al.,
features were reported in (Osadchiy et al., 2018) for user 2019) introduced the NutRec model, which specifies the
recommendations. The system uses association rules and healthiest ingredient and its dosage for recipes for nutritious
inverse filtering methods to recommend the results based foods. The framework takes into account quantity as well
on user item descriptors and preferences [1]. K- means as user-defined elements that are vital for a specific diet.
clustering and collaborative filtering were used by (Yuan et Results from experiments can demonstrate the superior
al., 2019) to execute diet advice [2]. The recipes in the healthfulness of recipes. The author of this paper uses the
cluster with higher ratings are those that best suit the user's data sets from Yummly and Allrecipes for improved
nutritional needs. The approach turned out to be 70% more outcomes [9]. The issue that occurs when conventional
precise and efficient. A recipe recommendation employing methods of recommendation are used to gather data from
ingredients based on user preference for similar foods was electronic health records was addressed by (Lore’s et al.,
proposed by (C. Yuen et al., 2012) [3]. 2012).
For hypertensive patients, (Sookrah et al., 2019) developed Therefore, the author developed a novel filtering technique
the DASH recommendation engine, which stands for that distinguishes between user and item features
dietary approaches to control hypertension and is based on individually, and the trial proved to be more helpful for
content-based filtering and machine learning [4]. Hybrid those with special health concerns and applicable to a
filtering was established to address the issues of comparable specific diet with quantity. The strategy was mostly focused
user behaviours and user interest in similar goods. To on consumers' interest in particular items of relevance to
address the cold start and sparsity issues, (Kim et al., 2020) them. The issue of sparsity, latency, and inappropriate
proposed a knowledge-based hybrid decision model that handling of persons with various needs and interests is
takes into account both physical and mental health. To resolved by this unique technique. For recipe
assist users who made decisions about their diet and recommendation, the author additionally uses a variety of
nutrition on a health platform, a food recommendation nature-inspired algorithms. (Singh et al., 2019) introduced
model was proposed [4]. By examining user demands, (Li the use of fuzzy logic and the analytical hierarchy method
et al., 2016) suggested a nutritional combination model. As to prescribe diets for people with diseases. They
more individuals switch to purchasing food online, the concentrated on the issue of undernourishment and the
quality and nutrients of the food are being sacrificed. The condition known as marasmus, which is brought on by a
model eliminates the cold start issue and provides high deficiency in nutrients such as carbs, fats, proteins, lipids,
performance since it is based on the functioning of glycogen, etc. [8]. (Singh et al., 2019) employed a fuzzy
comparable user behavior and user desire for item features. technique to prescribe the right diet, taking into account a
When a product is brand-new, the rating may be restricted variety of human factors based on the age group. The
since so many authors have employed knowledge-based method was employed since the diet data is quite
filtering [6]. After viewing the user's pathological reports, ambiguous and extensive.
(Rehman et al., 2017) proposed cloud-based food
recommendation, where ant colony optimization, a In this case, the author used a sizable data set in which age
technique inspired by nature, is used to generate an optimal group and nutrition were supplied into a fuzzy ontology as
food list and food recommendation, and accuracy is inputs, and food and nutrition were produced as a nutrition
increased by increasing the number of ants [7]. For %. According to the test results, the advised diet is more

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SILCHAR. Downloaded on August 30,2024 at 06:46:16 UTC from IEEE Xplore. Restrictions apply.
 accurate than the diet a dietician has recommended [7]. The proposed system makes improved recommendations and
expert recommendation method for the optimum nutrition suggestions to the user by utilizing a hybrid paradigm.
was proposed by (Chen et al., 2018). For the
recommendation of a tailored diet, a person's genetic III. PROPOSED FRAMEWORK
background is occasionally necessary. As it is necessary for Due to the coronavirus, people are now more aware of their
a tailored diet, people used to test their tests to obtain nutritional demands and watchful of their health [9]. The
precise information about their genes. The author has necessity of the hour is to maintain a healthy and nourishing
created a system to correlate a person's genotypic data with diet, and everyone is searching for the nutritional
data on supermarket items [6]. There are many different advantages of a specific item and how much of it they need
systems for making suggestions, therefore it was necessary for their own bodies. In this study, the author employed
to use a variety of innovative ways to improve the classifiers and machine learning algorithms to prescribe
effectiveness of the systems that represent the foods that diets based on user needs and requirements.
patients should eat. Many authors have employed machine
learning and deep learning since some specialists have
discovered that a balanced diet is an excellent choice to treat
patients with various diseases. Proposed IoMT-assisted
patient diet recommendation (Iwendi et al., 2020). For this
analysis, the author employed LSTM and demonstrated that
it had more accuracy. Random forest classifier was also
used. In this study, a sizable data set was used, and it was
split into training and testing data sets, respectively, with a
70:30 ratio. The health of the community is a vital concern,
although dietary requirements of an individual cannot
always be taken into consideration [5]. proposed food
recommendation system taking user preferences and
nutrition into account (Toledo et al., 2019). AHP sort and a
decision table were employed by the author. Twenty
nutrient food profiles and 600 food items are included in
their paper. There is a paucity of consumer food history in
this work, which could be exploited to improve the
Fig. 1 Proposed framework for nutrition-based diet recommendation
recommendations [6]. A model for diet advice was
proposed in 2020 by (V. Nallarasan et al., 2020) that is
based on the input parameters and recommends the diet The user will enter their age, height, weight, and
food. The authors' diet recommendation was based on a preferred diet (vegetarian vs. non-vegetarian). The diet
fuzzy inference-based method [7]. suggestion system examines the foods in each category
Some publications and research projects deal with making first, and then makes predictions based on the user's health
dietary and food recommendations for people based on their needs. Depending on the user's needs, the system will
work cultures and how sensors record their daily activities. suggest the required list of food products, regardless of
The consumer can receive product and service whether the user is healthy, overweight, or underweight. As
recommendations from web-based systems based on their seen in fig. 1, the recommended list of foods will be
preferences [9]. Similar to this, the suggested system provided for breakfast, lunch, and dinner. The suggested
follows this methodology when advising users on their diet framework demonstrates how the recommendation model
and food choices. The automated method for providing operates. Machine learning algorithms for clustering and
users with customized meals has been mentioned by George classification are implemented in Python and used to
Salloum et al. [2]. The authors covered patient data recommend a user's diet. The user enters the necessary
including dietary preferences, food variety, food-meal information, including age, weight, height, and vegetarian
compatibility, and inter-food compatibility. Another paper and non-vegetarian food preferences. A user's body mass
[7] provided further evidence of the work based on the index determines whether they are overweight,
consumption of tea and related beverages and how this underweight, or healthy. Food products are grouped into
affects the user's health. The study [4] illustrates the model breakfast, lunch, and supper clusters using K-means
to magnify the health problems and the suggested diet. clustering so that the user can receive more personalized
recommendations. Multiple attributes are inputted into the
To support the suggested system, numerous current models K-means clustering algorithm to create clusters. Based on
and methodologies are discussed in the text above. The the user-inputted values that fall into one of the clusters, the

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SILCHAR. Downloaded on August 30,2024 at 06:46:16 UTC from IEEE Xplore. Restrictions apply.
 classification method is utilized to forecast the list of food The sample data is filled and shown in figure 4. The output
items. of the input values is shown in figure 5.

The dietary items are grouped based on the

nutrients that are good for healthy, underweight, and
overweight people. For the purpose of recommending foods
that are suitable for a certain health status, the author used
the random forest classifier, support vector machine,
AdaBoost, and Gradient Boost. Through the Python-
designed Tkinter, the user will enter information.
Depending on the user's health status—overweight,
underweight, or healthy—a list of food products will be
suggested when the user enters his or her information.
Figure 2 depicts the workflow for the process.

Fig. 3 User Details

Fig. 4 Diet Recommendation input for Over-Weight Person

Fig. 2 Process flow diagram

IV. EXPERIMENTAL SETUP AND RESULT

ANALYSIS
Python 3's Scikit-Learn machine learning library is utilized
to construct this framework. The GUI interface utilizes
Tkinter. The most effective and reliable library for machine
learning is Scikit Learn. A variety of machine learning
algorithms are offered by Scikit Learn in the supervised and
unsupervised categories. Additionally, Scikit learn libraries
are used for classifiers. Accuracy, precision, recall, F1-
measures, and confusion matrices are some of the
performance features used to evaluate classifier
performance. The food data set is sourced from Kaggle and
includes nutrient values for calories, fat, and carbohydrates,
among other things. The values are the input for the
suggested framework, and Patient Age is the patient's age. Figure 5. Diet Recommendation output for Over-Weight Person
Patient Diet is used to specify the value of the vegetarian
and vegan diet, either 0 or 1. The patient's weight in kg. The Another sample data for the healthy diet plan is filled and
suggested diet will appear on the screen once the values shown in figure 6. The output of the input values for the
have been input. In figure 3, the output screen is displayed. healthy person is shown in figure 7.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SILCHAR. Downloaded on August 30,2024 at 06:46:16 UTC from IEEE Xplore. Restrictions apply.
 Table1: Recall and F1-measure values of proposed system

Performance Recall F1-measures

0 1 0 1
Classifier

Random Forest 0.71 0.88 0.71 .88

Support Vector 0.14 1 0.25 .84

Machine

AdaBoost 0.71 0.94 0.77 .91

Gradient Boost 0.71 0.88 0.71 .88

Fig. 6 Diet Recommendation for Healthy Person

V. CONCLUSION
In the proposed study, a diet advice system based on user
input values is demonstrated. Given everyone's hectic
schedules in today's world, it might be challenging to plan
or follow a diet based on one's health. The study discusses
a diet prediction model and strategy for this aim. The author
has outlined a framework for diet advice based on the user's
needs, their weight status—overweight, underweight, or
healthy—and two sample scenarios—a healthy person and
an overweight person. K-means clustering and classifiers
like Random Forest, Support Vector Machine, AdaBoost,
and Gradient Boost are utilised for recommendations. The
correctness was checked using the performance measures
F-measures, Recall, Accuracy, and Precision.
REFERENCES
[1] T. Osadchiy, I.Poliakov, P.Olivier, M.Rowland, and E.Foster.
"Recommender system based on pairwise association rules," Expert
Systems with Applications , July-2018.
Fig. 7 Diet Recommendation for Healthy Person [2] Z.Yuan, F.Luo, “Personalized Diet Recommendation Based on K-
means and Collaborative Filtering Algorithm,” Journal of Physics.
Performance Evaluation 2019.
[3] C. Yuen, Y.RU. Lin, L.A. Adamic, “Recipe recommendation using
ingredient networks,” 4th annual ACM web science conference, pp.
The performance of the proposed work is measured on 298-307, June 2012.
various performance indexes as shown in table 1 terms of, [4] R.Sookrah, J. Devee Dhowtal, S.Devi Nagowah, “A DASH Diet
recall and f1-measure. Recommendation System for Hypertensive Patients Using Machine
Learning ,” 7th International Conference on Information and
Table1: Accuracy and precision values of proposed system Communication Technology (ICoICT), pp. 1-6, July 2019.
[5] Z. Zhang, T.Peng, K.Shen, “Overview of Collaborative Filtering
Recommendation Algorithms,” IOP conference series earth and
Performance Accuracy Precision
environmental sciences, Vol. 440, 2019.
[6] C.Li, C.Yang, “The Research and Design of Recommendation
0 1 System for Nutritional Combo,” International Conference on
Computer and Communications (ICCC), pp. 830-837, May 2016.
Classifier
[7] H. Khatter, S. Arif, U. Singh, S. Mathur and S. Jain, "Product
Recommendation System for E-Commerce using Collaborative
Random Forest 0.82 0.71 0.88 Filtering and Textual Clustering," Third International Conference
on Inventive Research in Computing Applications (ICIRCA), pp.
Support Vector 0.73 1 0.73 612-618, 2021.
Machine [8] Harsh Khatter, Anil Kumar Ahlawat, “Analysis of Content Curation
Algorithms on Personalized Web Searching”, Proceedings of the
AdaBoost 0.86 0.83 0.88 International Conference on Innovative Computing &
Communications (ICICC), March 29, 2020.
[9] Jain A. Singhal, “Personalized Food Recommendation – State of Art
Gradient Boost 0.82 0.71 0.88 and Review,” Ambient Communications and Computer Systems,
Lecture Notes in Neworks and Systems, vol. 356, Springer,
Singapore.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SILCHAR. Downloaded on August 30,2024 at 06:46:16 UTC from IEEE Xplore. Restrictions apply.