Mini Review

A Deep Fuzzy Neural Network System to Group Moroccan Foods: Towards a Personalized Menu for Type 2 Diabetes Patient

El Moutaouakil Karim^1*, Roudani Mohammed¹, Ahourag Abdellah¹, Aayah Hammoumi², Mouna Cheggour³, Saliha Chellak³ and Hicham Baizri⁴

¹Engineering Science Laboratory (LSI), Polydisplinary Faculty of Taza, USMBA, Morocco

²Laboratory of pharmacology, neurobiology, anthropobiology and environment, Faculty of Sciences Semlalia, Cadi Ayyad University, Marrakesh 40000, Morocco

³Morphoscience laboratory, Faculty of Medicine and Pharmacy, Cadi Ayyad University, Marrakech, Morocco

⁴Biosciences and Health Research Laboratory, Avicenna Military Hospital, Cadi Ayyad University, Marrakech, Morocco

Submission: February 20, 2024; Published: July 23, 2024

*Corresponding author: El Moutaouakil Karim Engineering Science Laboratory (LSI), Polydisplinary Faculty of Taza, USMBA, Morocco

How to cite this article:El Moutaouakil K, Roudani M, Ahourag A, Aayah H, Mouna C, et al. A Deep Fuzzy Neural Network System to Group Moroccan Foods: Towards a Personalized Menu for Type 2 Diabetes Patient. Nutri Food Sci Int J. 2024. 13(2):555860. DOI: 10.19080/NFSIJ.2024.13.555860.

Abstract

Grouping foods is a very hard task because of the noncorrelation between nutrients. The methods based on a single feature led to restrictive groups. In this work, we introduce a grouping strategy that implements deep neural networks and fuzzy C-means. First, we collected the main information on the most consumed foods in Morocco based on 21 nutrients. Second, we use FCM to group similar nutrients. Third, we use the auto-encoder neural network to produce artificial foods by projection of the foods via the nutrients of the same groups. Fourth, we use FCM to decompose the set of foods into an appropriate number of groups. Two data transformations were performed when realizing our system: nutrients grouping and foods projection. In this context, two performance measures are adopted: Mean Square Error (MSE) and Silhouette (S). In this context, our system produces a homogenous group of foods.

Keywords:Deep Neural Network; Fuzzy C-means; Diabete; Diet

Introduction

The main importance of food grouping is to build a flexible diet so that patients maintain their diet over a long period of time to avoid the complex stages of certain chronic diseases such as diabetes. Several research teams have proposed a classification of vegetables by color. Vegetables of identical color have the same nutritional properties. For example, the food guides established by the United States Department of Agriculture (USDA) incorporate the seven fundamental food categories. While the USDA food directories cover all seven basic food categories, the MyPyramid (CNPP 2008) proposes four categories of basic food groups [1-8]. Vegetables provide a high source of vitamins (A, B, C, E), which are vital for health. The body derives numerous vitamins through food. It’s advisable to consume a diverse diet, full of a range of foods, in to ensure sufficient vitamins to sustain good health. There are many studies on vitamins and the risk of bladder.

Authors Suppressed Due to Excessive Length cancer (suggested by the World Cancer Research Fund and the American Institute for Cancer Research), and others on vitamin-based regimens for long-term illness, as likewise numerous studies on the contribution of vitamins to good health (vitamin D and bone health) [9,10]. The botanical categorization of plants is guided by physiology, i.e. the features of plant growth, structure, and organization. The utility of botanical categorization to satisfy the requirements of nutritionists is further complicated by the possibility that foods within the same botanical family might or might not include comparable degrees of nutrients. In addition, certain vegetables may be divided into more than one category when multiple parts of the vegetable are found in vegetable. category created when more than one portion of the plant is considered edible. A further difficulty is to categorize fruits and vegetables, in situations plants are consumed individually. Botanical categorization deals with the entire plant and is not limited to the various parts of the plant that are typically eaten. Finally, we point out that most groups established in literature are based on only one feature. Regrettably, these categories are not homogenous enough, as a single attribute is not sufficient to precisely identify foods [8]. In this work, we introduce a grouping strategy that implements deep neural networks and fuzzy C-means. First, we collected the main information on the most consumed foods in Morocco based on 21 nutrients. Second, we use FCM to group similar nutrients. Third, we use the auto-encoder neural network to produce artificial foods by projection of the foods via the nutrients of the same groups. Fourth, we use FCM to decompose the set of foods into an appropriate number of groups. Two data transformations were performed when realizing our system: nutrients grouping and foods projection. In this context, two performance measures are adopted: Mean Square Error (MSE) and Silhouette (S). In this context, our system produces a homogenous group of foods. The rest of this paper is structured as follows: Section 2 presents the main steps of the proposed method. Section 3 gives the experimental results and discussion. Section 4 concerns some conclusions and future directions.

Proposed System

In this section, we give different steps to realize our system, especially data collection, dimension reduction, artificial nutrients building, and foods grouping. At the end of this section, we give the used performance measures to evaluate the components of our system.

Data Collection: first, we collected the main information on the most consumed foods in Morocco based on 21 nutrients (micro and macro nutriments and glycemic load).

Dimension Reduction

 In this step, we use FCM to group similar nutrients. In this case, the number of samples is 21 (number of nutrients) and the number of features is 170 (number of foods) [5]. Unlike the hard methods, this method permits the objects to be in different groups, at the same time, using members. Title Suppressed Due to Excessive Length 3 ship functions [6]. To this end, the Fuzzy C-Means try to solve the following optimization problem:

where

is the ith sample from

informs us how much the sample zi is in the group c and w_c is the center of the c^th cluster. Fuzzy C-Means process in iterative optimization of the problem (FP).

Artificial Nutrients Building

In this step, we use the auto-encoder neural network to produce artificial foods by projection of the foods via the nutrients of the same groups [1-3]. The number of auto-encoders equals the number of groups containing more than one nutrient. The auto-encoder is a deep neural network composed of two principal sections: the encoder (box of neurons) and the decoder (box of neurons). The hidden layer gives the coded information, and the last layer must produce the input tests. The loss function E quantifies the sum of the local loosed information when transforming t to artificial tweet :

If the encoding operation is realized by a mapping P and D then

The global error is given by &

is the set of the collected tweets.

Foods Grouping

In this step, we use FCM to decompose the set of foods into an appropriate number of groups [4,6,7].

Performance Measure

Two data transformations were performed when realizing our system: nutrients grouping and foods projection. In this context, two performance measures are adopted: Mean Square Error (MSE) and Silhouette (S). Mean Square Error (MSE): if  and

are the original and the estimated sample, respectively, then the mean square error is defined by:

Silhouette: Suppose the data were divided into K groups by any technique, including GMM, fuzzy K-means, K-medoids or K-means.

It should be noted that the larger the silhouette, the more similar the data is to the group to which it was assigned. 4 Authors Suppressed Due to Excessive Length

Experiment Results and Setup

First, we use FCM to group the nutriments into 8 groups (this number is experimentally chosen). (Table 1) gives the obtained groups. We remark that only cluster 3 contains several (13) nutriments, especially vitamins. In addition, the Glycemic load falls within this group. Second, the foods are projected following each nutrient group using the auto- encoder neural network. Hidden Size=1’MaxEpochs’=3000 ’Encoder TransferFunction’=’satlin’;’DecoderTransferFunction’=’purelin’;L2WeightRegularization’=0.01;’SparsityRegularizati’SparsityProportion’=0.10.’Sparsity Proportion’=0.10. The mean square error associated with this projection is 81.29 micrograms (Figure 1 & Figure 2 & Figure 3). Third, we use FCM to group the foods into 8 groups (this number is experimentally chosen). (Table 2) gives the food groups silhouette. We remark that all the groups are strongly correlated except the third one. (Table 2) gives the obtained groups. We remark that groups 3 and 6 contain the largest number of foods, which offer a high diversity to the patient’s foods. To compare the groups obtained with our system to the ones produced with FCM, directly, we use this later to group the foods. (Figure 4) gives the groups of foods silhouette produced with FCM directly applied to the foods (53.9142). We remark that the silhouette of this latter is less than the one of the groups obtained with our system. Because the clustering methods are incapable of grouping the data and understanding the correlations between features.

Conclusion

In this work, we propose a deep neural network and fuzzy C-means to decompose foods into homogeneous groups. Our method processed into four steps (a) col- lection of the main information on the most consumed foods in Morocco based on 21 nutrients, (b) grouping of the nutriments using FCM, (c) projection of the foods using the auto-encoder neural network, (d) grouping of the foods using fuzzy C-means. The performance of our system was evaluated based on MSE and silhouette. The error of the projection is 81.29 mg, and the quality of the groups is 53.9142mg. As a consequence, our system produces homogenous groups of foods. In future work, we will use our system to build personalized diets for health conditions people, especially diabetic people.

Acknowledgment

This work was supported by the Ministry of National Education, Professional Training,

Higher Education and Scientific Research (MENFPESRS) and the Digital Development Agency (DDA) of Morocco (Nos. Alkhawarizmi/ 2020/23).

References

1. Tchessalov S, Maglio V, Kazarin P (2023) Practical Advice on Scientific Design of Freeze-Drying Process: 2023 Update. Pharm Res 40: 2433–2455.
2. K Nakagawa, A Horie, M Nakabayashi, K Nishimura, T Yasunobu (2021) Influence of processing conditions of atmospheric freeze-drying/low-temperature drying on the drying kinetics of sliced fruits and their vitamin C retention,Journal of Agriculture and Food Research 6: 100231.
3. Ajani CK, Zhu Z, Sun DW (2023) Shrinkage during vacuum cooling of porous foods: Conjugate mechanistic modelling and experimental validation. Journal of Food Engineering 337: 111220.
4. Rahman MS (2009) Food Stability Beyond Water Activity and Glass Transition: Macro-Micro Region Concept in the State Diagram. International Journal of Food Properties 12(4): 726-740.
5. Claussen IC, Andresen T, Eikevik TM, Strømmen I (2007) Atmospheric Freeze Drying—Modeling and Simulation of a Tunnel Dryer. Drying Technology 25(12): 1959-1965.
6. Claussen IC, Strømmen I, Hemmingsen AKT, Rustad T (2007) Relationship of Product Structure, Sorption Characteristics, and Freezing Point of Atmospheric Freeze-Dried Foods. Drying Technology 25(5): 853-865.
7. González-Fésler M, Salvatori D, Gómez P, Alzamora SM (2008) Convective air drying of apples as affected by blanching and calcium impregnation. Journal of Food Engineering 87(3): 323-332.
8. Nieto A, Salvatori D, Castro MA, Alzamora SM (1998) Air drying behaviour of apples as affected by blanching and glucose impregnation. Journal of Food Engineering 36(1): 63-79.
9. Lewis WK (1921) The Rate of Drying of Solid Materials. Journal of Industrial & Engineering Chemistry 13: 427-432.
10. Crank J (1975) The Mathematics of Diffusion. 2nd Edition, Oxford University Press, London. Pp. 69-88.
11. Tolstorebrov I, Eikevik TM, Bantle M (2014) A DSC study of phase transition in muscle and oil of the main commercial fish species from the North-Atlantic. Food Research International 55: 303-310.
12. Riedel L (1978) Eine Formel zur Berechnung der Enthalpie Fettarmer Lebensmittel in Abhangigkeit von Wassergehalt und Temperatur. Food Chemistry, microbiology, technology 5(5): 129-133.
13. Gordon M, Taylor JS (1952) Ideal copolymers and the second-order transitions of synthetic rubbers in non-crystalline copolymers. Journal of Applied Chemistry 2(9): 493-500.
14. Champion D, Le Meste M, Simatos D (2000) Towards an improved understanding of glass transition and relaxations in foods: molecular mobility in the glass transition range Trends in Food Science and Technology 11(2): 41-55.
15. Truong V, Bhandari BR, Howes T, Adhikari B (2004) Glass transition behaviour of fructose. International Journal of Food Science & Technology 39(5): 569-578.
16. Mrad ND, Bonazzi C, Boudhrioua N, Kechaou N, Courtois F (2012) Moisture Sorption Isotherms, Thermodynamic Properties, and Glass Transition of Pears and Apples. Drying Technology 30(13): 1397-1406.
17. Shafiur Rahman M (2007) Glass Transition and State Diagram of Foods Handbook of Food Preservation, Second Edition: CRC Press. pp. 335-363.
18. Riedel L (1956) Kaltetechnik. 8: 374.
19. Claussen IC, Ustad TS, Strømmen I, Walde PM (2007) Atmospheric Freeze Drying-A Review. Drying Technology 25(6): 947-957.
20. Li Y, Somorjai GA (2007) Surface Premelting of Ice. The Journal of Physical Chemistry C 111(27): 9631-9637.
21. Duan X, Ding L, Ren Gy, Liu Ll, Kong Qz (2013) The drying strategy of atmospheric freeze-drying apple cubes based on glass transition. Food and Bioproducts Processing, 91(4): 534-538.
22. Champion D, Le Meste M, Simatos D (2000) Towards an improved understanding of glass transition and relaxations in foods: molecular mobility in the glass transition range Trends in Food Science and Technology 11(2): 41-55.
23. Bantle M, Kolsaker K, Eikevik TM (2011) Modification of the Weibull Distribution for Modeling Atmospheric Freeze-Drying of Food. Drying Technology 29(10): 1161-1169.
24. Petrova I, Eikevik TM (2016) Drying Kinetics of Pork Muscle Obtained for Five Different Drying Modes and Three Salt Concentrations. In 20^th International Drying Symposium (IDS). Gifu, Japan.
25. Doymaz I (2009) An Experimental Study on Drying of Green Apples. Drying Technology 27(3): 478-485.
26. Golestani R, Raisi A, Aroujalian A (2013) Mathematical Modeling on Air Drying of Apples Considering Shrinkage and Variable Diffusion Coefficient. Drying Technology 31(1): 40-51.
27. Veltchev ZN, Menkov ND (2000) Desorption Isotherms of Apples at Several Temperatures. Drying Technology 18(4-5): 1127-1137.
28. Quintas M, Brandão TRS, Silva CLM, Cunha RL (2006) Rheology of Supersaturated Sucrose Solutions. Journal of Food Engineering 77(4): 844-852.
29. Zogzas NP, Maroulis ZB, Marinos-Kouris D (1996) Moisture Diffusivity Data Compilation in Foodstuffs. Drying Technology 14(10): 2225-2253.
30. Vega-Gálvez A, Ah-Hen K, Chacana M, Vergara J, Martínez-Monzó J, et al. (2012) Effect of temperature and air velocity on drying kinetics, antioxidant capacity, total phenolic content, colour, texture and microstructure of apple (var. Granny Smith) slices. Food Chemistry 132(1): 51-59.
31. Kaymak-Ertekin F, Gedik A (2004) Sorption isotherms and isosteric heat of sorption for grapes, apricots, apples and potatoes. LWT - Food Science and Technology 37(4): 429-438.
32. Sá MM, Figueiredo AM, Sereno AM (1999) Glass transitions and state diagrams for fresh and processed apple. Thermochimica Acta 329(1): 31-38.