Nanotechnology Perceptions

ISSN 1660-6795
www.nano-ntp.com

Analysis Bidding Price Analysis Using K-

Means Cluster, Machine Learning, And Deep
Learning in Public Institution Competitive
Bidding: Targeting Service Projects in The
Chungcheong Region
Young-Hun Kim1, Gitae Kim2
1
Department Of Smart Production & Management Engineering, Hanbat National
University, Daejeon, Republic of Korea
2
professor, Department of Industrial & Management Engineering, Hanbat National
University, Daejeon, Republic of Korea

The probability of participating in a public institution bidding, which is a burden of expenditure,

from winning the bid to signing a contract is quite low. In addition, because the conclusion of a
contract has a significant impact on the company's management, bidders are under a lot of stress
regarding bidding. Therefore, in this paper, a successful bid price analysis study was conducted to
enable bidders to easily predict the bid price. Clusters were analyzed using the big data analysis
algorithm K-means Cluster, Decision Tree Regressor, Random Forest Regressor, Support Vector
Regressor, and K-Nearest. The accuracy of the results was analyzed using Neighbors Regressor,
Multi-layer Perceptron Regressor, etc. To this end, a price-to-value analysis was performed based
on the last three years of service bidding results data in the Chungcheong region.

Keywords: Bidding price, Competitive bidding, Clustering, Machine learning, Deep learning

1. Introduction
With the introduction of the e-Government system in the 2000s, electronic bidding began
through the Public Procurement Service. When public institutions make announcements
through competitive bidding in accordance with contract laws, they must announce
competitive bidding through the Public Procurement Service's National Marketplace. In order
to participate in competitive bidding, two of the 15 multiple reserve prices consisting of +7
and -8 are selected, and when the bidding process is conducted, the expected price is
determined by taking the arithmetic average of the four most frequently selected multiple
reserve prices. Before the 2000s, information related to bidding was obtained through public

Nanotechnology Perceptions 20 No. S4 (2024) 157–166

 Analysis Bidding Price… Young-Hun Kim et al. 158

institution bidding information sheets, but since the 2000s, information related to bidding has
been collected from the Public Procurement Service or bidding-related sites. Services provided
by bidding-related sites include information on the Public Procurement Service opening bid
results, information on the number of companies participating in the bidding, information on
the expected price, information on multiple reserve prices, and data on past bidding results.
Based on the information provided, bidding participants predict the bidding price by selecting
a section where the expected price is high, selecting a section where the number of bidders is
low, and selecting a section where the expected price is high and where there are few bidders.
Kang Min-seok (2014), who studied the impact of the multiple reserve price creation range
and multiple reserve price section setting method on the scheduled price based on competitive
bidding standards, discussed the multiple reserve price creation range, multiple reserve price
creation section, and multiple reserve price lottery method, studied[1]. In this study, it was
found that in determining the expected price for a successful bid, it was most reasonable to
select the multiple reserve price lottery method as the top two and the bottom two as the
unequal division method. Kim Cheol (2017) studied the factors affecting the public IT project
success rate prediction model and used linear regression, support vector machine regression,
and random forest in an R program to create a success rate prediction model using public IT
success characteristic data Developed [2]. In this study, variables that affect the bid success
rate were identified and a plan for utilizing them for public IT bidding was presented.
Daeseong Hwang (2020) studied bidding price prediction using deep learning in electronic
bidding using bidding result data for electrical construction using Machine Learning and Deep
Learning methods [3]. In this study, training and test data were split 7:3, and the prediction
accuracy in the Deep Learning algorithm was high. Previous papers are papers on the creation
range of multiple reserve prices, multiple reserve price creation sections, and the method of
setting multiple reserve prices to the top two and bottom two, and bidding using bidding
information data from the information and communication industry with a machine learning
algorithm to Price prediction accuracy was analyzed. And using the bidding result data for
electrical work, the accuracy of the bidding results was predicted using Machine Learnig and
Deep Learning methods. This study used bid result data for the service industry to analyze
similar characteristics by dividing them into the same clusters using K-mearns Cluster, and
analyzed the accuracy of successful bid price prediction for the service industry using Machine
Learnig and Deep Learning. It can be said that it is different from existing previous papers.
Projects ordered by public institutions have good cash flow, so there are many businesses
participating in competitive bidding, and the probability of participating in the bidding and
winning the bid is quite low. Additionally, the performance of the bidding has a significant
impact on the company's management. As a result, the company's management puts
psychological pressure on the bidding performance on the bidding manager, and the burden of
bidding develops into stress on the bidding manager. For this study, 32,813 bid result data for
service industries ordered in the Chungcheong region were collected from January 1, 2019 to
October 15, 2023. This data was preprocessed and K-means clustering was performed on
18,441 data. Similar characteristics of the data were analyzed using an algorithm and divided
into 9 clusters. And with 17,925 pieces of data, the accuracy of bid price prediction was
analyzed using Decision Tree Regressor, Random Forest Regressor, Support Vector
Regressor, K-Nearst Neighbors Regressor, and Multi-layer Perceptron Regressor. As a result,
Random Forest Regressor was the most accurate, with an accuracy of over 99%. As a result

Nanotechnology Perceptions Vol. 20 No. S4 (2024)

159 Young-Hun Kim et al. Analysis Bidding Price....

of this study, the company's management status will be improved and the bidding manager's
work stress regarding bidding will be reduced.

2. Literature Review
2.1 Public bidding
In order to ensure fairness, bids that impose a burden on financial expenditure are subject to
competitive bidding in accordance with the National Contract Act and the Local Contract Act.
At this time, from bid notice to bidding and contract, you must go through the competitive
bidding process in accordance with the National Contract Act or Local Contract Act. 15
multiple preliminary prices must be prepared and posted at least 5 days before the opening
date in accordance with the procedures set forth in the government contract regulations,
“Standards for Preparing Prices” and “Standards for Local Government Bidding and Contract
Execution,” and businesses participating in the bidding must submit them before the opening
date. You must participate in the bidding after meeting the qualifications according to the
relevant laws and regulations[4][5]. To participate in the bidding, register on the Public
Procurement Service's National Marketplace, access the virtualized site of the National
Marketplace, and select two out of 15 multiple reserve prices. The ordering organization
registers as a user on the Nara Marketplace and then proceeds to bid. At this time, the expected
amount is calculated by taking the arithmetic average of the four most frequent multiple
reserve prices. And after the opening of bids is completed, priority in qualification review will
be given to the business that participated in the bidding closest to the scheduled price. The
contract partner will be determined after screening the first-ranked company in accordance
with the Ministry of the Interior and Safety’s contract regulations, “Standards for determining
successful bidders when bidding for local governments”[6][7]. When calculating 4 out of 15
multiple reserve prices, the expected price is 1365, and the formula is as in (1).
𝑛𝑃𝑟
𝑛𝐶𝑟 = − − − − − − − − − − − − − − − − − − − − − − − −(1)
𝑟!
The lower limit of the expected price in competitive bidding is limited. In order to minimize
problems caused by poor design due to receiving low-priced orders, the lower limit of the
expected price is applied at the time of bidding, and the formula is as follows (2) and (3).
88 𝐵𝑖𝑑 𝑃𝑟𝑖𝑐𝑒
𝐺𝑟𝑎𝑑𝑒 = 80 − 20 | ( − ) × 100 | − − − − − − − − − (2)
100 𝐸𝑥𝑝𝑒𝑐𝑡𝑒𝑑 𝑃𝑟𝑖𝑐𝑒
88 𝐵𝑖𝑑 𝑃𝑟𝑖𝑐𝑒
𝐺𝑟𝑎𝑑𝑒 = 90 − 20 | ( − ) × 100 | − − − − − − − − − (3)
100 𝐸𝑥𝑝𝑒𝑐𝑡𝑒𝑑 𝑃𝑟𝑖𝑐𝑒
Businesses participating in competitive bidding must prepare the limit amount according to
formula (2) in the case of a service bid with a basic amount of less than 100 million won among
1,365 scheduled prices, and (3) in the case of a service bid with a basic amount of less than
200 million won. Prepare a bid with the limited amount according to the number formula and
submit the bid through a cadastral information processing device.
2.2 K-means Clustering
Unsupervised learning is used to explain the distribution of data, extract key features of data,
Nanotechnology Perceptions Vol. 20 No. S4 (2024)
 Analysis Bidding Price… Young-Hun Kim et al. 160

or model the distribution of data. Representative methods include principal component

analysis (PCA), linear discriminant analysis (LDA), and clustering. The second is a method of
creating a new representation of data, and representative methods include artificial neural
network (ANN), One-Hot Encoding, and Text Summarization [8][9]. K-means clustering, a
type of cluster analysis, classifies the acquired data into k clusters. Each cluster is defined
based on the similarity of the data, and the data within the clusters are similar to each other,
and the data between the clusters are not similar to each other [10]. Groups are formed based
on data characteristics, clustered based on the central point of the group, and data with similar
characteristics are grouped to help recognize patterns [11].
2.3 Machine Learning Model and Deep Learnning Model
Decision Tree Regressor, a type of machine learning, is an algorithm that can predict the final
dependent variable by composing a complex decision-making process into a combination of
several simple decisions. It is a non-parametric model used for both regression and
classification [12]. Random Forest was proposed by Breiman in 2001, and is a model that
generates bootstrap sampling based on a decision tree and trains the same algorithm multiple
times by extracting multiple subsets from the entire training dataset [13]. Support Vector
Regressor is a type of classification algorithm that is a generalized model of Support Vector
Machine. It is a model that predicts data by finding the optimal hyperplane that includes as
much data as possible within the distance between support vectors [14],[15]. K-Nearest
Neighbors Regressor is a model that operates by checking k Lables located at the closest
distance from the value Fast and easy to use [16],[17]. Multi-layer Perceptron Regressor is a
type of Deep Learning. It is a feedforward neural network with one or more hidden layers
between the input layer and the output layer. The hidden layer is an acceptable continuous
function and is a model that is advantageous for finding approximations and calculating
weights [18],[19][20][21][21][22][23].

3. Proposed Work
When 32,813 data from three years of service industry bidding results in the Chungcheong
region were preprocessed and 18,441 data were analyzed using K-means clustering, they were
clustered into 9 groups as shown in Table 1.
cluster 1 2 3 4 5 6 7 8 9 Sum
Count 968 3,032 3,435 3,409 2,341 1,707 1,228 1,152 1,169 18,441
Table 1. K-means clustering data grouping
Table 2 shows the distribution range, average, and number of data of the expected price by
amount, and the ratio of dividing the lowest price for the range by group was recorded in the
expected price range. It was found that the scheduled price was distributed within a range of
±3% according to the government's contract regulations. Since the range from 10 million won
to 20 million won is generally conducted through voluntary bidding, contracts are concluded
in the range of 90% of the basic amount. Therefore, the expected price range was found to be
relatively high. The average range of groups 2 to 9 is 0.8783 to 0.8768, and it can be seen that
the expected price is slightly biased toward the negative sign.

Nanotechnology Perceptions Vol. 20 No. S4 (2024)

 161 Young-Hun Kim et al. Analysis Bidding Price....

Division amount(won) range average Count Expected price range lower limit price

1 group 10,000,000~20,000,000 0.87378~0.90356 0.88841 968 99.2931~102.6773 88%

2 group 20,000,000~30,000,000 0.86998~0.88688 0.87830 3032 98.8613~100.7818 88%

3 group 30,000,000~40,000,000 0.86984~0.88595 0.87798 3435 98.8454~100.6761 88%

4 group 40,000,000~50,000,000 0.86992~0.88596 0.87793 3409 98.8545~100.6773 88%

5 group 50,000,000~60,000,000 0.86819~0.88548 0.87730 2341 98.6579~100.6227 88%

6 group 60,000,000~70,000,000 0.86710~0.88504 0.87689 1707 98.5340~100.5727 88%

7 group 70,000,000~80,000,000 0.86563~0.88460 0.87621 1228 98.3670~100.5227 88%

8 group 80,000,000~90,000,000 0.86651~0.88502 0.87664 1152 98.4670~100.5705 88%

9 group 90,000,000~100,000,000 0.86705~0.88497 0.87689 1199 98.5284~100.5648 88%

Table 2. Expected price range, average, and lowest bid price for each group
Fig 2. is data visualizing the analysis results of the expected price by group. In Group 1,
contracts were often carried out through private contracts or private bidding, and the number
of companies that could participate in bidding was limited, resulting in a high expected price.
It can be seen that groups 2 to 9 are widely distributed within a limited range. This indicates
that the expected price shows a normal distribution. In order to apply an algorithm for
predicting the bid price, it is possible to analyze the trend of the expected price distribution
according to the probability distribution, and the section of the expected price that was widely
distributed in the past can be analyzed recently. If the distribution is low at the time, it can be
predicted that the probability of the scheduled price appearing in the range of that group will
increase.
Fig 2. Visualization of expected price analysis results by group

Table. 3 shows an example of preprocessed data of competitive bidding results.

Basic Expected EP/BA EP/BA 1st priority 1st priority 1st place basic Number of
Data preparation companies
amount price (100%) (0%) Bid amount Assessment rate
5163 32,066,000 31,9015,900 99.5319 -0.4680 28,099,530 99.5798 87.6302 12
21251 74,015,000 73,398,150 99.1665 -0.8334 64,637,299 99.2386 87.3299 144
3965 51,664,000 52,212,850 101.0623 1.0623 46,548,000 102.3835 90.0975 3
21234 41,800,000 41,411,850 99.0714 -0.9285 36,452,940 99.0999 87.2079 46
19470 50,000,000 49,0971,250 98.1425 -1.8575 43,677,800 99.2677 87.3556 17

Tabe 4. includes 9 variables (columns) of 17,925 pre-processed bid result data, and the real
number data (float64) of each variable is divided into estimated price, base price, scheduled
price, expected price/basic price (100%), Analysis of trends in bidding price changes in the
Chungcheong region using variables such as expected price/basic price (0%), 1st priority
Nanotechnology Perceptions Vol. 20 No. S4 (2024)
 Analysis Bidding Price… Young-Hun Kim et al. 162

bidding amount, 1st priority evaluation rate (100%), 1st priority basis comparison, and number
of companies, estimated price and basic price , This is information that analyzes the
comparative analysis of the expected price, the relationship between the first bid amount and
<class ‘pandas.frame.dataFrame’>
Int64Index: 17925 entries, 5163 to 15795
Data columns (total 9 columns):
# Column Non-Null Count Dtype
--- ----------- ------------------- ---------
0 Estimated price 17925 non-null float64
1 Basic amount 17925 non-null float64
2 Expected price 17925 non-null float64
3 EP/BA(100%) 17925 non-null float64
4 EP/BA(0%) 17925 non-null float64
5 1st priority Bid amount 17925 non-null float64
6 1st priority Assessemnt rate 17925 non-null float64
7 1st place basic preparation 17925 non-null float64
8 Number of companies 17925 non-null float64
Dtypes: float64(9)
Momory usage: 1.4MB

Train_size : test_size = 8 : 2
Table 4. Data Information
the basic price, and the relationship between the number of participating companies and the
bid price. To prevent overfitting of the model and improve generalization performance, the
data split ratio was set at 80% of the total data as training data and 20% as test data. Table 5
shows the analysis results for each model, and the Random Forest model showed the highest
accuracy and generalization performance. The Decision Tree model showed high accuracy but
had a possibility of overfitting, the SVR model had a high possibility of underfitting, and the
KNN model was sensitive to data size. MLP required a lot of data and learning time to increase
data accuracy. It was analyzed that the Random Forest Model, which generally shows high
accuracy and generalization performance, is most suitable for the learning purpose of
predicting the characteristics of bidding result data and bidding price.
Study Explanation
Model MSE R2 Score
Time(ms)
Decision High accuracy (R2 Score), but possible
95949447797.46814 0.999745 309.757
Tree overfitting
Random Higher accuracy and generalization performance
67120411441.58901 0.999822 259,076
Forest than DT
389033901300897.4 Sensitive to data distribution and possible
SVR -0.032547 -
4 underfitting
230768234718.2910 Sensitive to data size and high computational
KNN 0.999388 -
2 cost
Requires a lot of data and learning time for high
MLP 396074769078.2721 0.998949 480,383
accuracy
Table 5. Data example
When performing Parameter Grid Search with Random Forest Model, the combination of
max_depth=8, min_samples_leaf=8, min_samples_Split=8, and n_estimators=100 was
selected as the optimal hyperparameter. The prediction accuracy was very high at 0.9997 based
on R2 Score when optimal hyperparameters were used. The RMSE of the training data was

Nanotechnology Perceptions Vol. 20 No. S4 (2024)

163 Young-Hun Kim et al. Analysis Bidding Price....

251,927.41, and the RMSE of the test data was 305,923.92. There is a possibility of overfitting
of the RMSE of the test data, but it was not considered a major problem because the prediction
accuracy of the model was very high. < Table 6. Reference >. Fig 3 and Fig 4 are the learning
curves of training and test data RMSE.

# Parameter definition for grid search

Param_grid = {
‘n_estimators’: [10, 50, 100, 200],
‘max_depth’: [None, 5, 8, 10, 20],
‘min_samples_split’: [2, 5, 8, 10],
‘min_sam;le_leaf’: [1, 2, 4, 8, 16]
}
Opimal parameters:
{‘max_depth’: 8 ‘min_samples_left’: 8 ‘min_samples-split’ : 8 ‘n_estimators’: 100}
Optimal prediction accuracy: 0.9997

Training data RMSE : 251927.41295753248

Test data RMSE : 305923.9244689635

Table 6. Random Forest Grid Search

Fig 3. Trannig Data learning curve

Nanotechnology Perceptions Vol. 20 No. S4 (2024)

 Analysis Bidding Price… Young-Hun Kim et al. 164

Fig 4. Test Data learning curve

4. Conclusion
This study used bidding result data from the Chungcheong region, analyzed the similarity of
the data into 9 groups using K-means Cluster, analyzed it into 9 clusters, and visualized it.
Through this cluster, it was found that the minimum floor rate of the bid price varies depending
on the basic amount. In addition, in order to predict the accuracy of bidding results and bidding
price, the accuracy of bidding results and analysis is measured using Decision Tree Regressor,
Random Forest Regressor, Support Vector Regressor, K-Nearst Neighbors Regressor which
are types of machine learning, and the accuracy of bidding results and analysis if measured
using Multi-layer Perceptron Regressor which is a type of Deep Learning was analysis. As a
result, Random Forest (RF) had a high prediction accuracy of 99.97%, and the test data RMSE
was 305,923.92 and the training data RMSE was 251,927.41. The test data RMSE is higher
than the training data RMSE, so there is a slight possibility of overfitting, but the prediction
accuracy is very high, so there is a large possibility of overfitting is It was analyzed as not a
problem. We preprocessed 32,183 bid result data from the Chungcheong region from January
1, 2019 to October 15, 2023, and used 17,925 bid result data as actual data to create predicted
values trained using the expected price (actual value) and training data. When comparing the
bid prices, it was confirmed that the values were almost close to the successful bid price.
Although the actual and predicted values did not completely match, it was found that almost
similar graphs appeared. Therefore, among the prices predicted in the experiment, there are
many cases that fall between the successful bid price and the lower limit price, so it is analyzed
that it is possible to predict the bid price with a high probability of winning when predicting
the bid price using the Random Forest algorithm. The results of this study can be expected to
increase the probability of winning a bid when predicting the bidding price using the Random
Forest algorithm, and are expected to have a positive impact on alleviating the psychological
stress of bidders and the company's business performance. Considering that 17,925 pieces of
Nanotechnology Perceptions Vol. 20 No. S4 (2024)
165 Young-Hun Kim et al. Analysis Bidding Price....

data were used as actual data when 32,183 pieces of data were preprocessed, it appears that a
lot of unnecessary data was included in the data collection process. This can be seen as
insufficient classification of unnecessary data during the data collection process, and it can
affect more precise analysis, so it was necessary to collect more precise data during the data
collection process. In addition, when analyzing Deep Learning with an algorithm, it was
affected by computer performance, so the limitations of the research cost burden and analysis
time were felt. The next study is expected to increase the accuracy of prediction by adding an
algorithm using Deep Learnig, and conduct experiments to obtain the optimal hidden layer or
node by changing the parameter values in more diverse ways, which will allow analysis to the
winning bid price. The next study uses nationwide service bidding data to increase the
accuracy of prediction by adding an algorithm using Deep Learnig, and experiments to obtain
the optimal hidden layer or node by changing the parameter values in more diverse ways will
bring the result closer to bid price.

References
1. M.S.Kang, “The effect of multiple reserve price generation range and section setting method
on estimated price in bidding and successful bidding system”, Mater’s degree thesis at Korea
University Graduate School of Management and Information, pp. 11-15, 43, 2014.
2. C. Kim, “Development of bidding-ratio prediction model for public information technoloty
business projects using data mining method,” Master’s degree thesis at Yonsei University
Graduate School of Department of Inustrial Information Management, pp. 15-20, 2017.
3. D. H. Hwang, Y. C. Bae. (2020). Bidding price prediction using deep learning in electronic
bidding. Journal of the Electronics and Telecommunications Society of Korea, 15(1), 147-
152.
4. Ministry of Strategy and Finance established regulations No. 653, 2023.06.16, partially
revised, (contract regulations) standards for preparing expected prices.
5. Ministry of the Interior and Safety Regulation No. 252, June 29, 2023, partially revised, local
government bidding and contract execution standards.
6. Ministry of Strategy and Finance, Contract Regulations, 2022.06.01. Partial revision, full
text of contract regulations.
7. Ministry of the Interior and Safety Regulation No. 253, June 29, 2023, partially revised,
criteria for determining successful bidders when bidding for local governments.
8. Tom M,Mitchell, Introduction to Machine Learning, 1997
9. Michael I.Jordan, Unsupervised Learning:A Theoretical and Empirical Analysis, 2002
10. wikipedia, K-means clustering, https://en.wikipedia.org/wiki/K-means_clustering
11. A. Goia, C. May, and G. Fusai, “Functional clustering and linear regression for peak load
forecation”, Elsevier, International Journal of Forecation”, Elsevier, International Journal of
Forecastion, Vol, 26, No. 4, pp. 700-711, Oct, 2010.
12. Cha Ki-wook, Hong Won-hwa. (2023). Development of a prediction model for the amount
of decommissioned waste generated using a decision tree-based algorithm. Journal of the
Architectural Institute of Korea, 39(3), 179-187.
13. Breiman, L. (2001). Random forests. Machine learning, 45(1), 5-32
14. Jeong Sang-geun, Bae Jin-ho, Kang Hye-gyeom, and Kim Seong-yeol. (Date). Development
of SVR-based short-term power demand prediction algorithm using public data. Korean
Institute of Electrical Engineers conference proceedings, venue.
15. Taeyoung Kim and Jinho Kim, “Solar power generation prediction using support vector
regression and Python library”, Proceedings of the Korean Institute of Electrical Engineers

Nanotechnology Perceptions Vol. 20 No. S4 (2024)

 Analysis Bidding Price… Young-Hun Kim et al. 166

conference, pp. 540-541, 2018

16. Seon-ho Kim, Nak-hoon Choi, Jong-seok Oh. (2023). Dissolution prediction study of anti-
tuberculosis drug capsules using lNN and ensemble. Journal of the Korean Society of
Industrial-Academic Technology, 24(1), 531-537, 10.5762/KAIS.2023.24.1.531.
17. Z. Zhang, “Introduction to machine learning: k-nearest neigbors”, Annals of translational
medicine, vol. 4, no. 11, pp. 218, Jun. 2016. DOI:https://doi.org/10.21037/atm.2016.03.37
18. Su-han Kim, Ho-rim Wang, Won-ju Lee, Byeong-hyuk Ahn, Yu-jeong Kim, Joo-ho Lee, and
Hyun-seong Shin. (2023). Molecular dynamics data-based hyperelastic constitutive equation
modeling using MLP, GR, and RBF artificial neural networks. Journal of the Korean Society
of Mechanical Engineers, Volume A, 47(1), 49-57, 10.3795/KSME-A.2023.47.1.049.
19. Adnaa, J. Daud, N. G. N., Ishak, M. T., Rizman, Z. I. and Rahman, M. I. A., 2018, “Tansig
Activation Function (of MLP network) for Cardiac Abnormality Detection,” AIP Conference
Proceedings, Vol. 1930, No. 1, Article 02006.
20. Se-Joon Park,Yung-Cheol Byun, Electric Mobility Demand Prediction by Region using K-
means Clustering”, Journal of KIIT. Vol. 19, No. 11. Pp. 125-132, Nov. 30. 2021.
http://dx.doi.org/10.14801/jkiit.2021.19.11.125
21. Maltare, N. N., Sharma, D., Patel, S. (2023). An Exploration and Prediction of Rainfall and
Groundwater Level for the District of Banaskantha, Gujrat, India. International Journal of
Environmental Sciences, 9 (1), 1-17 https://www.theaspd.com/resources/v9-1-1-
Nilesh%20N.%20Maltare.pdf
22. Min, P.K., Mito, K. and Kim, T.H. (2024). The Evolving Landscape of Artificial Intelligence
Applications in Animal Health. Indian Journal of Animal Research.
https://doi.org/10.18805/IJAR.BF-1742
23. Kim, T. H. and AlZubi, A.A. (2024). AI Enhanced Precision Irrigation in Legume Farming:
Optimizing Water Use Efficiency. Legume Research. https://doi.org/10.18805/LRF-791

Nanotechnology Perceptions Vol. 20 No. S4 (2024)