论文信息
论文解读
AI 辅助整理- 方法
- Kaggle水果营养数据集,直方图均衡化预处理,贝叶斯多层感知机分类。
- 发现
- BMLP在准确率等指标上表现优异,能处理营养模式不确定性,优于传统方法。
- 机会点
- 可结合真实膳食与产地数据,构建面向推荐系统的可解释营养分类模型。
论文摘要
机器学习研究主要集中于水果识别和病害识别能力,而关于营养成分分类的研究较少。传统的水果分类方法基于水果的物理特征,主要是大小、形状、颜色和质地,而较少重视营养成分组成。本研究旨在提出一种智能且自动化的基于营养成分的水果分类方法,以辅助健康饮食决策、精准农业和水果推荐系统。本研究主要关注利用贝叶斯多层感知机(Bayesian Multilayer Perceptron, BMLP)模型对水果营养成分聚合体进行分类,该模型有望提高基于营养成分的水果分类的准确性和效率。所开发的系统从Kaggle数据集仓库获取水果数据集,其中包括维生素、矿物质、碳水化合物、蛋白质、脂肪、膳食纤维、水分和能量值。随后对数据进行直方图均衡化(Histogram Equalization)预处理,以增强营养成分含量并提取相关特征。提取的特征用于通过贝叶斯多层感知机进行水果营养成分聚合体分类。为衡量模型的有效性,采用准确率(Accuracy)、精确率(Precision)、召回率(Recall)和F1分数(F1-score)进行评价。结果表明,本研究采用的机器学习方法取得了准确的性能表现。模型中引入了贝叶斯概率理论,在提高分类准确性和良好处理营养成分模式不确定性方面表现优异,并优于传统分类技术。
全文
FUDMA Journal of SciencesFruit Nutrient- Based Classification … Lawal-Habeebullahi and Abdulrauf FJS
FUDMA Journal of Sciences (FJS) Vol. 10 No. 24, December, 2026, pp 271 – 275
271
Fruit Nutrient-Based Classification of Selected Food Crops using Bayesian Multilayer
Perceptron Algorithm
*Kafayah Idowu Lawal-Habeebullahi and Tosho Usman Abdulrauf
1Department of Computer Science. Faculty of Computing, Engineering and Technology, Al-Hikmah University, Ilorin,
Kwara State, Nigeria.
*Corresponding authors’ email: mykafayahabdullahi@gmail.com
ABSTRACT
Received:
Accepted:
Published:
25 Sep. 2026
11 Nov. 2026
15 Nov. 2026
Research on machine learning has focused primarily on the ability to recognise fruit and
identify diseases, but there has been little work done on the classification of nutrients.
The traditional approaches to fruit classification are based on physical fea tures of fruit,
primarily size, shape, color and texture, and little emphasis is placed on nutritional
composition. The study is intended to: present an intelligent and automated nutrient-based
fruit classification approach that will aid in healthy dietary decisions, precision
agriculture and fruits recommendation systems. The main concerns of this study are the
classification of fruit nutrient aggregates by utilizing a Bayesian Multilayer Perceptron
(BMLP) model which is expected to enhance the accuracy an d efficiency of fruit
classification based on nutrients. The system developed gets the fruits datasets from the
kaggle datasets repository, which includes vitamin, minerals, carbohydrate, protein, fat,
dietary fibre, moisture and energy values. It then wil l be subjected to a pre -processing
called Histogram Equalization to enhance nutrients contents and extract relevant features.
The extracted features are used for fruit nutrient aggregates classification by using a
Bayesian Multilayer Perceptron. To measure the effectiveness of the model, it is
evaluated in terms of Accuracy, Precision, Recall, and F1 -score. The result yielded an
accurate performance to the adopted Machine learning used on this research. Bayesian
probability theory is incorporated into the model, which is performed excellently in terms
of improving the classification accuracy and handling the uncertainty of nutrient patterns
well and outperforms conventional classification techniques.
Keywords:
Fruits classification, Nutrients, aggregates, Human health, Bayesian Multilayer
Perceptron
INTRODUCTION
Agriculture is the most important crop production system in
the world and fruits and vegetables is a source of essential
nutrients needed for good health. It provides vitamins that no
living being can survive without. Since both humans and
animals depend on crop every single day, there must be a high
value on classifying these crops into various classes. There is
strong evidence that fruits are packed with vital nutrients and
natural compounds that, when eaten in the right amounts, it
provides incredible health benefits to the body. Just as certain
"friendly" microbes help the gut, the vitamins, minerals and
fiber found in fresh fruit can be classify into various class of
mineral contents. (Espinoza et al, 2023).
According to research by Hassan, et al (2025) which show
that eating a wide variety of these crops can significantly
boost the immune system and improve metabolic health. This
supports the idea that our bodies are designed to thrive on
natural, plant -based foods. The vitamins provided by fruit
nutrients help our immune systems stay sharp and responsive.
By regularly eating fruit, it can improve our heath and help
our bodies fight off modern lifestyle health issues that are
often caused by poor diets. Healt h problems such as chronic
inflammation, heart issues and skin problems can often be
managed or even reversed - by the natural healing properties
found in fruit. Essentially, the nutrients in fruit act as a natural
defense system, keeping our bodies balanced and healthy in a
world full of processed foods.
Rushith and Rohith (2023), stated that crops are good for
human health, also riched in vitamin A, vitamin B, vitamin C,
various inorganic salts, calcium, phosphorus, iron, iodine,
etc., and contains a variety of amino acids, which are all
necessary nutrients for the human body. Fruits and vegetables
are indispensable in daily life. With the development of
economy and the im provement of consumption level,
consumers' demand for high -quality fruits is gradually
increasing. During fruit ripening, fruit quality and vitamins
are mainly determined by the presence or absence of crop
defects and nutrients. At present, fruit nutrients aggregates
classification needs to be done on selected types of fruits
Apple, Banana and Cabbage. In addition, due to the
advancement of agricultural technology, multiple varieties of
each fruit are bred with different characteristics to meet the
heath needs. Therefore, there is need to accurately classify
each crop into their nutrient type and vitamin contents.
It is not just about the quantity of the harvest, but also the
quality and the nutritional value of what is being produced.
However, a major challenge facing the agricultural sector
today is the difficulty in accurately classifying and grouping
the nutrients found in various crops, especially fruits. This is
essential to a healthy lifestyle and are consumed daily by
people all over the globe. Mukhriddin, et al, (2022), Several
fruits are packed with vital ingredients that act as natural
medicine for the body. Crop that are not properly classify or
understand these nutrients, becomes difficult to use or prevent
specific health issues. Many common ailments such as the flu,
persistent headaches, the common cold, stomach ache, and
even eye conditions can be improved by eating the right kinds
vitamins found in fruit.
FUDMA Journal of Sciences (FJS)
ISSN online: 2616-1370
ISSN print: 2645 - 2944
Vol. 10 No. 24, December, 2026, pp 271 – 275
DOI: https://doi.org/10.33003/fjs-2026-1024-6121
Fruit Nutrient- Based Classification … Lawal-Habeebullahi and Abdulrauf FJS
FUDMA Journal of Sciences (FJS) Vol. 10 No. 24, December, 2026, pp 271 – 275
272
Jongen (2025), further discussed, fruits play a massive role in
skin health, helping to fix dryness, prevent early aging, and
clear up skin discoloration. Hence, research focuses on using
Multilayer Perceptron (MLP), which is a powerful type of
Artificial Neural Network (ANN) in classifying the selected
fruits (Apple, Banana and Cabbage). This model is designed
to act like a digital brain that can learn com plex, non-linear
patterns that simple computers might miss. The MLP is
highly effective at classification, regression, and prediction
tasks, making it the perfect tool for analyzing the selected
fruits (Apple, cabbage and Banana) sets. By training the
network on specific physical features, the system can
accurately categorize different types of fruit and predict their
nutrient levels at high speeds and a smarter way to detect and
classify these beneficial fruits. Because classification cannot
be done manually, the use of Multilayer perceptron (MLP) to
analyze fruit types and their healing properties. This research
highlights how specific fruits, when eaten in certain amounts,
can target and provide the needed vitamins to the body. By
using these modern technical methods, it provides an
automated solution that makes it easier to understand fruit
nutrients. The model approaches not only help the agricultural
industry but also offers a natural path to better health by
showing exactly how different fruits can be used as effective
to provide the right class of nutrients contents. Many journals
have image recognition and fruits diseases classification but
little studies had on fruits nutrients aggregates and
classification.
Lawal (2023), stated the process of classifying fruits entails
grouping into categories according to a variety of vitamins.
These criteria can include, but are not limited to, size, shape,
color, flavor, and botanical features. This classification can
serve several functions, such as those in agriculture, botany,
nutrition, and the culinary arts. The use of Multilayer
perceptron for the classification of fruits offers several
significant applications and benefits across a variety of fields.
Models that utilize Artificial Neutral Network can be utilized
to assist in the evaluation of the nutrients of fruits based on
characteristics such as size, shape, color, and flaws. The use
of this information can help farmers know and group the fruits
according to their vitamin contents. The estimation of crop
yields that can be obtained by fruit nutrients classification can
be of use in the health management and Fruit aggregates.
Kumar, et al, (2023) considered the amount of effort that has
been invested in classifying biologically active components
within fruits. In another paper by Jongen (2025), who stated
that rich source of a variety of vitamins, minerals, dietary
fiber and ma ny other classes of bioactive compounds
collectively called phytochemicals. In the natural
environment, there are many kinds of fruits, and there are
more and more high-quality varieties. Consumers have more
choices of high -quality varieties, and different varieties also
determine nutrient content. The traditional classification of
fruit varieties is based on the shape, color, surface texture,
size and other characteristics of the fruit. Some fruit varieties
are relatively similar in maturity, size, color and taste, and the
accuracy of classification by traditional methods. The
solution is to automate the classification process using
Multilayer perceptron.
Ragini et al, (2020) stated that consumption of fruits in day -
to-day dietary plays an important role in the nutritional
supplement intake of humans. Estimation of right nutrient
intake through fruits consumption has become increasingly
significant in order to maintain the proper health of the
wellbeing. Salem (2024), Fruit is a major food products in
their own right and key ingredients in many processed foods.
Consumers increasingly require food products that preserve
their nutritional value, retain a nat ural and fresh colour,
flavour and texture, and contain fewer additives such as
preservatives.
MATERIALS AND METHODS
This presents the methodology adopted for the classification
of fruit nutrient aggregates using a hybrid approach that
integrates Multilayer Perceptron (MLP) with Bayesian
techniques for feature optimization. Figure 1 below display
the system architecture.
Figure 1: Workflow of the fruit classification
and detection system
Dataset Acquisition
The dataset was extracted from Kaggle repository
(https://www.kaggle.com/moltean/fruits), the dataset
contains 360 different images of the three crop and fruit
engaged in this research. The data were splited into 70% for
training and 30% for testing.
Data Pre-processing
The acquired data are rarely ready for direct use, so it passes
through a pre -processing stage. Typical operations include
resizing images to a uniform dimension, noise removal,
colour space conversion, normalisation of pixel values and
where necessary, da ta augmentation through rotation,
Fruit Nutrient- Based Classification … Lawal-Habeebullahi and Abdulrauf FJS
FUDMA Journal of Sciences (FJS) Vol. 10 No. 24, December, 2026, pp 271 – 275
273
flipping or scaling. Pre -processing improves the consistency
of the data, reduces the influence of irrelevant variations and
helps the model learn more efficiently.
Feature Selection
Following pre -processing, the feature selection stage
identifies the attributes that best distinguish one fruit from
another, such as colour, shape, size and texture. By retaining
only the most informative features and discarding redundant
or irrelevant on es, this stage reduces dimensionality, lowers
computational cost and lessens the risk of overfitting. The
resulting feature set is then divided into two branches, as
shown in the figure 1: the training data that contain 70% and
the testing data with the remaining 30%.
Training and Testing Data
Splitting the data into training and testing subsets allows the
model to be built on one portion of the data and judged on
another that it has not seen. The training data is used to teach
the model the relationship between the extracted features and
the fruit classes. The testing data is held back to provide an
unbiased measure of how the model performs on new
samples. This separation is important for detecting overfitting
and for confirming that the model can generalise beyond the
examples it learned from.
Train Model and Model Evaluation
In the Train Model block, the selected classification algorithm
learns the patterns that characterise each fruit category. The
Model Evaluation block then assesses the trained model using
performance measures such as accuracy, precision, recall and
F1-score, often supported by a confusion matrix. As depicted
in the figure, the evaluation result feeds back into the training
stage. If performance is unsatisfactory, parameters can be
tuned, features revised or the training data adjusted, and the
model is retrained until acceptable results are obtained. This
feedback loop m akes the workflow iterative rather than
strictly linear.
Fruit Classification and Detection
Once the model has reached satisfactory performance, it is
used for fruit classification, in which an input image is
assigned to one of the learned fruit categories. The final stage,
detection, applies the classification outcome to identify and
report the fruit present in the image. This output is the
practical result of the system and can support applications
such as automated sorting, grading, quality control and
inventory management in agriculture and retail.
RESULTS AND DISCUSSION
In order to generate reproducible results, the fruit -nutrient
classification framework was evaluated within a controlled
and standardized computing environment. The specific
configuration used for the experiment is summarized below.
Table 1: Statistical Properties of preprocessed Feature Vector
Feature Apple Banana Cabbage
y1 (Red Intensity) 0.78 ± 0.05 0.82 ± 0.06 0.31 ± 0.05
y2 (Green Intensity) 0.15 ± 0.03 0.79 ± 0.05 0.78 ± 0.04
y5 (Aspect Ratio) 0.98 ± 0.02 0.31 ± 0.08 0.95 ± 0.03
y6 (Circularity) 0.94 ± 0.03 0.42 ± 0.07 0.42 ± 0.07
y10 (Uniformity) 0.83 ± 0.05 0.86 ± 0.04 0.58 ± 0.06
Table 2: Nutrient Comparison across the Three fruits (per 100 grams)
Nutrient Apple Banana Cabbage
Calories (kcal) 52 89 25
Protein (g) 0.26 1.09 1.28
Fibre (g) 2.4 2.6 2.5
Vitamin C (mg) 4.6 8.7 36.6
Potassium (mg) 107 358 170
Table 3: Comparative Algorithm Performance Matrix
Method Accuracy Precision Recall F1-Score
Bayesian MLP
(New Model)
95.77% 95.55% 95.27% 95.37%
Standard MLP 91.20% 90.85% 91.00% 90.92%
Support Vector Machine 88.50% 88.12% 88.30% 88.21%
K-Nearest Neighbors (KNN) 84.10% 83.90% 84.00% 83.95%
Discussion
The feature means are presented in the Table 2, along with the
range of values, for each class, for one normalised feature (all
values are in the range [0, 1]). The three types of information
retained are: colour (y1 red intensity, y2 green intensity),
shape (y5 aspect ratio, y6 circularity) and surface appearance
(y10 uniformity). The numbering (y1, y2, y5, y6, y10)
indicates that these are a carefully selected group of features
that would be part of a larger feature vector, suggesting that
only most infor mative and/or interpretable features are
reported.
i. Red intensity (y1). The highs are Apple (0.78) and banana
(0.82) while the much lower is cabbage (0.31). The means of
these two classes differ by only 0.04, less than either class
spread; this is the feature that distinguishes cabbage from the
other two classes, but not apple from banana.
ii. Green intensity (y2). Apple is very low but banana (0.79)
and cabbage (0.78) are almost the same and high. This feature
is used to isolate apple. Combined with y1, the pattern
corresponds to RGB colour mixing such that high red and low
green results in a red apple, high red and high green results in
a yellow banana, and low red and high green results in a green
cabbage.
Fruit Nutrient- Based Classification … Lawal-Habeebullahi and Abdulrauf FJS
FUDMA Journal of Sciences (FJS) Vol. 10 No. 24, December, 2026, pp 271 – 275
274
iii. Aspect ratio (y5). Banana is strongly elongated (0.31)
while apple (0.98) and cabbage (0.95) are close to being as
broad as they are high. This feature is the banana separated
out from the rest of the image and has the highest variance of
any of the featur es (0.08 ±) which indicates how curved and
the direction of the bananas in an image vary.
iv. Circularity (y6). Apple is nearly perfect (0.94), banana is
0.42 and cabbage is 0.42. As such, the feature can distinguish
apple from the other fruits but cannot distinguish banana from
cabbage. The low value is expected for banana and plausible
only for c abbage (see Section 4) when the leafy, irregular
outline increases the perimeter.
v. Uniformity (y10). Apple (0.83) and banana (0.86) have
smooth and even surfaces, while the cabbage (0.58) is clearly
lower, alluding to its veined and layered leaf surface. Like the
red intensity, this can differentiate cabbage, but not apple
from banana.
The worth of that identification can be seen in Table 3, as
once an item is recognised its nutrient profile can be looked
up. Banana has a different nutritional profile, cabbage has a
low calorie profile, but is high in vitamin C and fibre and
apple has a humble profile on most nutrients. The most
energy-dense banana has an energy density of 89 kcal; about
1.7 times higher than apple (52 kcal) and 3.6 times higher than
cabbage (25 kcal). As expected, the lightest of them all is
cabbage because it contains the most water. All three items
contain low amounts of protein. Cabbage (1.28 g) and banana
(1.09 g) are about 4 -5 times higher than apple (0.26 g) but
none are significant sources of protein by weight, at about
1%.
The three items are close to each other in fibre (2.4 to 2.6 g)
with a difference of 0.2 g which is not significant in practice.
A big difference in energy terms is covered up by a similarity
in mass terms. Cabbage is about 20 times as fibre dense as
apple (0.10 g fibre/100 kcal) and more than 30 times as fibre
dense as banana (0.029 g fibre/100 kcal).
The greatest differences and opposite trends occur between
the two micronutrients. Cabbage has the highest quantity of
Vitamin C (36.6 mg) which is more than 8 times the amount
in the apple (4.6 mg) and more than 4 times the amount in the
banana (8.7 mg). The recommended reference values for
cabbage vary according to country, age and sex, but 100 g
consumed by an adult will contribute about 40 to 50 % of the
average daily requirement for an adult (see reference values).
Banana has the highest potassium content (358 mg) which is
approximately 3.3 times as high as that of apple (107 mg) and
2.1 times as high as cabbage (170 mg), but is still around a
tenth of the typical daily adequate intakes. It can be seen in
this table that, among the 3, Apple has the least protein, fibre,
vitamin C and potassium in it whilst cabbage has the least
calories.
In Table 4, across every one of the four metrics reported, the
newly proposed Bayesian MLP outperforms all three baseline
algorithms and the size of this performance gap widens
noticeably as the comparison moves from the Standard MLP
down through the Suppo rt Vector Machine to the KNN
classifier. This descending pattern is consistent with the
general expectation that simpler, non -probabilistic models
tend to struggle more as feature overlap between classes
increases, whereas a model equipped to represent uncertainty
explicitly is better positioned to cope with that overlap. The
relative ordering of the baselines is also informative in its own
right. The Standard MLP, despite lacking the Bayesian
probability layer, still outperforms both the Support Vector
Machine and the KNN classifier, suggesting that a neural -
network architecture in general is well matched to this
particular feature set, and that the incremental gain from the
Bayesian formulation is additive on top of an already
reasonably capable base arch itecture rather than a substitute
for it. The Support Vector Machine's mid -table position is
broadly in line with its known strengths on moderately sized,
well-separated feature spaces, while the comparatively lower
accuracy of the KNN classifier is consistent with its general
sensitivity to local feature overlap and to the choice of
neighbourhood size, both of which become more
consequential as class boundaries blur, as appears to be the
case for some of the produce comparisons in this dataset.
Conclusion
The accuracy of Bayesian MLP (95.77%) and the precision,
recall and f1 -score of Bayesian MLP (95.55%, 95.27% and
95.37%) for classification of apple, banana and cabbage are
greater than the accuracy of the baselines (Standard MLP,
Support Vector Machine an d KNN) (91.72%, 92.23% and
89.72%) respectively, 92.23% and 89.51% respectively. The
explicit uncertainty modelling is a valuable architectural
component especially when there is a visual separation of the
fruit classes that is overlapping. Well distinguis hed by three
crops with their colour, shape and uniformity of surface and
a reference value can be used for retrieval of nutrient profile
(e.g. high level of vitamin C in cabbage, moderate level in
apple, high level of potassium in banana, and energy density).
This suggests applications in the field of dietetics such as
healthy food selection as well as recommendation systems for
fruit use and in agriculture such as automated sorting and
grading, quality control and precision agriculture. Some are
still potential uses since they depend upon images of the crops
and the comparison of nutrient information to that stored in
the system not the actual measurement of nutrient levels.
Moreover, the study was carried out on 3 crops and limited
number of images (360). Further testing of the model to larger
and more varied data sets and more crop varieties in real
world conditions in the farm, market and diet is then required
and the results of the model compared with the laboratory
nutrient analysis. Unless validated, the method will not be
useful for food choice and in crop management.
REFERENCES
Espinoza, S., Aguilera, C., Rojas, L. A., & Campos, P. G.
(2024). Analysis of fruit images with deep learning: A
systematic literature review and future directions. IEEE
Access, 12 , 3837 –3859.
https://doi.org/10.1109/ACCESS.2023.3345789
Hassan, M.A.F., Majeed, A.W. & Hasan, M.A. (2025). Fruits
and their role in boosting the immune system (review article,
published online 31 December 2025). This one addresses fruit
vitamins and phytochemicals in immune function. The
authors' names are in Arabic script in the source, so I
transliterated them.
Jongen, W. (Ed.). (2025). Fruit and vegetable processing:
Improving quality . Woodhead Publishing. 639 pages.
Business & Economics
Kaggle. (n.d.). Fruit 360 dataset [Data set].Retrieved from
https://www.kaggle.com/moltean/fruits
Kumar, S., Singh, A., & Sharma, R. (2023). Web -based
application for calorie estimation using machine learning and
image processing. Journal of Food Informatics , 8(3), 234 -
256.
Fruit Nutrient- Based Classification … Lawal-Habeebullahi and Abdulrauf FJS
FUDMA Journal of Sciences (FJS) Vol. 10 No. 24, December, 2026, pp 271 – 275
275
©2026 This is an Open Access article distributed under the terms of the Creative Commons Attribution 4.0
International license viewed via https://creativecommons.org/licenses/by/4.0/ which permits unrestricted use,
distribution, and reproduction in any medium, provided the original work is cited appropriately.
Lawal O. M. (2023), “Study on strawberry fruit detection
using lightweight algorithm,” Multimed. Tools Appl.,
https://doi.org/10.1007/s11042-023-16034-0
Majeed, A. W., Muneer, N. Y., & Hassan, M. A. F. (2025).
Fruits and their role in boosting the immune system: Article
review. Journal of Medicinal and Industrial Plant Sciences,
3(4), 37–51. https://doi.org/10.32894/MEDIP.25.3.4.4
Mukhiddinov, M., Muminov, A., & Cho, J. (2022). Improved
classification approach for fruits and vegetables freshness
based on deep learning. Sensors, 22 (21), Article 8192.
https://doi.org/10.3390/s22218192
Ragini M.1, D. Lavanya1, B. Likitha2, A. SivaRam3, and Ch.
Yesu Narayana 4 (2020), Vegetables with its Nutrients
Classification of Fruits and its nutrients.
Salem S. A,. Hassan N. A, and R. M. Hamdy, (2024), “Impact
of transfer learning compared to convolutional neural
networks on fruit detection,” J. Intell. Fuzzy Syst., 2024,
https://doi.org/10.3233/jifs-233514
论文原文与出版信息
FUDMA Journal of Sciences