International Journal of Advances in Engineering and Management (IJAEM)

Volume 6, Issue 02 Feb 2024, pp: 575-580 www.ijaem.net ISSN: 2395-5252

Smart Agriculture Prediction Awaring IoT

Using Recurrent Neural Network
1
1
Dr.R.Hemalatha, 2Ms.G.Nivashini
Research Supervisor, Head &Associate professor, PG & Research Department Of Computer Science, Tiruppur
Kumaran College For Women, Tiruppur, Tamilnadu, India
2
Research Scholar (Ph.D), PG & Research Department Of Computer Science,
Tiruppur Kumaran College For Women, Tiruppur, Tamilnadu, India
---------------------------------------------------------------------------------------------------------------------------------------
Date of Submission: 28-02-2024 Date of Acceptance: 08-03-2024
---------------------------------------------------------------------------------------------------------------------------------------
ABSTRACT: Mostly on agricultural research, this The problems made worse by climate
has a big impact on choices like pricing, change and the rising demand for food brought on
distribution, and import-export of particular by the world's fastest-growing population—which
commodities. Utilizing cutting-edge technologies is is predicted to exceed 9 billion people by 2030—
vital for enhancing yield quality and creation, underline the need of an accurate crop yield
projecting crop yields, and researching agricultural forecast model at the regional level. The model
illnesses and infections. The most common needs to be improved in terms of crop
problem with farmers is that they don't choose the management, food security, policy, and agricultural
right crop according to what the soil requires. They making choices. Because of this,
observe a notable decline in production as a result. numerous methods have been created and used to
In order to address the farmers' problem, we forecast agricultural yield, including process-based
proposed a deep reinforcement learning (DRL) models and data-driven statistical algorithms [2].
based crop categorization system for precision However, statistical methods offer intriguing
agricultural selection in this study. In the crop alternatives and supplementary tools, since the
recommendation system, DRL-based advanced process-based models for simulating the
agricultural approaches reduce unfavourable physiological mechanisms are limited by data
possibilities and increase output. We contrasted the availability for parameterization, model calibration,
suggested recommendation system with a number and validation. The development of big-data
of machine learning techniques, including Random technology and high-performance computing has
Tree, Naive Bayes, and K-Nearest Neighbor, for a led to the popularity of a useful statistical technique
site-specific crop with effective accuracy. called deep reinforcement learning (DRL). The
Keywords: Agriculture, deep neural network, IoT, DRL algorithm produced new possibilities to
recommendation system, KNN, Random Tree, support farmers' decision-making, and guide
Naive Bayes, reinforcement learning, SVM decisions in a range of real-world scenarios with or
without little assistance from humans. Using
I. INTRODUCTION datasets generated from various sources, DRL
Crop yields are mostly predicted using approaches preserve the benefit of autonomously
crops including wheat, rice, beans, pulses, sugar solving large nonlinear problems; they also offer a
cane, tea, cotton, greenhouses, corn, and soybeans. robust and flexible framework for decision-making
Forecasting crop yields is essential to the world's and integrating expert knowledge into the system
food production. Seed businesses forecast new based on data.
hybrids to be employed in varied soil conditions, so
they can create superior types for different II. LITERATURE REVIEW
environments. With the assistance of The agricultural accuracy of the dataset
Crop forecasts, farmers gain by preventing created for in [4], the forecast of soybean crop
monetary losses. Farmers need to know which production over several years is calculated by the
crops are appropriate for a given season and what authors to examine the effectiveness of
safety measures to take based on the soil and classification algorithms. The authors of this paper
environment conditions [1]. Farmers also need to use Bayes classification, random forests, neural
know what kind of crops to cultivate by looking at networks, and vector machines. Authors in [5]
previous years' yields. emphasized the significance of crop prediction, and

DOI: 10.35629/5252-0602575580 |Impact Factorvalue 6.18| ISO 9001: 2008 Certified Journal Page 575
 International Journal of Advances in Engineering and Management (IJAEM)
Volume 6, Issue 02 Feb 2024, pp: 575-580 www.ijaem.net ISSN: 2395-5252

it can support the development of an extendable Regression approaches were employed by

framework for agricultural yield forecast in policy- the authors in [12] to estimate yields in a specific
making. It enables the adaptable integration of site with satisfactory outcomes. This study has
multiple methods for yield prediction in demonstrated that agricultural yields within a
agriculture. Additionally, a tool that uses dependent certain geographic area may be accurately
and independent variables to assist users in estimated using the regression model. Uses
predicting agricultural productivity in various crops concepts from descriptive analytics to analyze
has been built. agricultural data [13]. This study offers information
The study [6] uses visual data mining to on which data analytics techniques are used on
illustrate the agricultural data. Large-scale sugarcane harvest datasets, contingent on the
agricultural data is gathered for this project in order investigation's findings. In [14], the ramifications
to gain understanding of the results and of their findings are examined along with the
applications of Large-scale agricultural data is consequences of using machine learning techniques
gathered for this study in order to gain to large datasets related to sugar cane production in
understanding of input application and output (such Karnataka, India. Many machine learning
as fertilizers). The dataset has been reduced by the techniques have been applied to agricultural data in
use of multidimensional scaling algorithms and an effort to determine which methodology is the
autonomous maps. The study in [7] highlights the most productive. Three different supervised
significance of crop selection and talks about the learning algorithms were used to examine the sugar
factors that affect it, including market pricing, cane dataset: DT, k-NN, and SVM. DRL
production rates, and government regulations. In techniques, which integrate the ideas of deep
order to solve the issue of crop selection and learning and reinforcement learning to enable
increase the crop's net return rate, this study artificial agents to directly acquire knowledge and
proposes the Crop Selection Method (CSM). This experience from real-world data, can be applied to
suggests that the plant sequences for a season, anomaly detection [15].
taking into account plant species, weather, soil
composition, and water density. Article [8] III. PARAMETERS
forecasts cereal production in Bangladesh's In agriculture, a number of factors,
principal districts using data mining techniques. To including temperature, soil, moisture, rainfall, and
compute agricultural productivity in different humidity, affect crop productivity. We look into
districts, the data set included two area-related how the aforementioned factors affect the
variables, three biotic variables, and five production of agricultural output. The variables are
environmental parameters. Future geospatial listed below.
analytical initiatives to increase accuracy were a) Temperature: Temperature is a major factor in
suggested in the paper. Several classification the crop recommendation approach. There is a
methods are offered by the research in [9] to maximum and minimum temperature range for
classify the data set for liver disease. It has been each crop. Low temperatures can harm crops by
shown that naive Bayes classifiers perform better in smothering, freezing, and chilling them. Elevated
predictions. The multi-layer perceptron is the temperatures may lead to problems in absorbing
method with the highest accuracy among the nutrients, disruption of photosynthetic processes,
others, mentioned. The issue of food insecurity in and death of plants.
Egypt is discussed in the article [10], which aids in
determining whether or not massive food imports b) Soil: When recommending crops, soil is a
are required. After analysis of the paper soil crucial factor. The soil contains the elements
datasets [11], categorization is anticipated where nitrogen, phosphorus, potassium, magnesium, and
The crop yield is expressed as a classification rule sulfur. The soil types that are classified by nutrients
based on the specific kind of soil. For predicting are Clay, Sandy, Salty, Peaty, Chalky, and Loamy.
agricultural productivity, the k-Nearest Neighbor
and Naive Bayes algorithms are employed. c) Moisture: The most important element in plant
Subsequent research endeavors have yielded development is the presence of moisture in the soil.
proficient models that employ several In the system, crop moisture recommendations are
categorization methodologies, such as supporting also crucial.
vector machines.

DOI: 10.35629/5252-0602575580 |Impact Factorvalue 6.18| ISO 9001: 2008 Certified Journal Page 576
 International Journal of Advances in Engineering and Management (IJAEM)
Volume 6, Issue 02 Feb 2024, pp: 575-580 www.ijaem.net ISSN: 2395-5252

Table 1: Analysis of A Dataset

P N K TEMPERAT HUMIDITY PH RAINFALL LABEL
URE
0 41 90 42 21.8797444 82.002744 6.502985 202.935536 RICE
1 58 84 40 22.7724652 81.320389 7.038096 226.655537 RICE
2 54 61 44 23.004459 82.320763 7.804027 263.964648 RICE
3 37 74 42 20.491096 80.158636 6.840207 248.864034 RICE
4 43 77 41 19.130175 81.604873 7.628473 262.713874 RICE

d) Precipitation: Rainfall is a major source of

water for a lot of crops. Harvesting of crops is 4.2.2 K-Nearest Neighbor
dependent on rainfall. In regions that receive K-Nearest Neighbor is determined using
enough rainfall, crops that require more water will both classification and regression. K-Nearest
be planted. Rainfall also makes a wider range of Neighbors is a simple technique that uses
water resources available, which makes it possible similitude measures to categorize new cases and
to grow a variety of crops. stores all of the examples that already exist. The
e) Humidity: Photosynthesis cannot take place sample set is divided into groups based on "close-
without humidity. Low humidity stunts the growth down" metrics, which include Manhattan and
of plants, whereas high humidity can degrade the Euclidean distances.
quality of plants.
4.2.3 Naive Bayes
IV. METHODOLOGY Based on the Bayesian classification
4.1 Data Sets Collection technique, the Naive Bayes classifier is a main
The data covers soil-specific properties probabilistic model that applies the Bayes theorem
that were acquired from the soil testing laboratory with naïve strong assumptions of autonomy. A
in Madurai District. Comparable online resources method for creating classification models called
for general agricultural information have also been Naive Bayes uses problem instances as vectors of
used. Plants that we grow include rice, millet, characteristic values, and a class label is given to
pulses, cotton, nuts, vegetables, sorghum, sugar every scenario. Class labels are selected from a
cane, and coriander. One of the primary elements limited set of possibilities. It is actually a group of
contained in the agricultural yield datasets shown algorithms built around the same fundamental idea
in Table 1 are used. rather than a single method. When the class
variable is present, all Bayes classifiers incorrectly
4.2 Learners Used In The Model assume that the value of a particular feature is
4.2.1 Random Tree independent of the significance of any other
A decision tree and a random tree are feature.
similar. As each split has only a random subgroup
of features, it differs from a random tree. It is i) Collection of data
possible to construct random trees with both ii) Pre-processing of data
nominal and numerical data. The Random Tree is iii)Divide the data into test data and train data
C4.5 or similar to CART. But during training, it iv) Apply training data set to the classifiers, KNN,
differs by selecting only a random subset of Random tree, naives bayes, DRL
characteristics. It considers K. Attributes chosen at v) Predict the accuracy of the yield prediction of
random for every node. The subset size is the crop
determined by the subset ratio argument. vi) Trained model and crop recommendation

DOI: 10.35629/5252-0602575580 |Impact Factorvalue 6.18| ISO 9001: 2008 Certified Journal Page 577
 International Journal of Advances in Engineering and Management (IJAEM)
Volume 6, Issue 02 Feb 2024, pp: 575-580 www.ijaem.net ISSN: 2395-5252

Figure 1: Flow Chart of The Proposed Algorithm

4.2.4 DRL learning methods is given, based on deep

Deep reinforcement learning is able to reinforcement learning. The environment evaluates
create an end-to-end architecture that can predict the categorization action the agent performs on a
the crop by combining the benefits of deep learning single sample at each time step and awards the
and reinforcement learning. Thus, a fresh agent. Given that the minority class sample offers a
intelligent method that can overcome the bigger reward, the agent responds to the minority
disadvantages of the previously mentioned machine class more strongly.

CLASSIFICATION ACCURACY
1.01 0.99644
1
0.99
0.97586
0.98
0.96588
0.97 0.95687
0.96
0.95
0.94 0.93564 CLASSIFICATION
0.93 ACCURACY
0.92
0.91
0.9
DECISION KNN RANDOM NAVIE DRL
TREE FOREST BAYES

FIGURE 2: ACCURACY OF VARIOUS CLASSIFIERS

DOI: 10.35629/5252-0602575580 |Impact Factorvalue 6.18| ISO 9001: 2008 Certified Journal Page 578
 International Journal of Advances in Engineering and Management (IJAEM)
Volume 6, Issue 02 Feb 2024, pp: 575-580 www.ijaem.net ISSN: 2395-5252

V. RESULTS AND DISCUSSION KNN, Naïve Bayes, and DRL based classifiers.
The crop can view different settings for Among the remainder, random forest and DRL
different parameters in any of these options at any networks based on neural networks perform the
one time. The system will be able to use these best.
possibilities to make decisions, which is why there Based on prior considerations and
are so many of them. In other words, every system discoveries, the current research may serve as
is sufficiently trained to function independently. inspiration for a number of future works. In order
Each crop's specific needs for nitrogen, phosphate, to mitigate the effects of over fitting, we first wish
and potassium are taken into consideration, along to widen the network structure. For instance, we
with temperature, humidity, rainfall, and crop type think that incorporating older data, like Long Short
labeling. After the entire data set has been Term Memory, into the new pipeline could
collected, the system must produce an input, which enhance the results. We also like to observe the
is determined by training. The system assigned all impact of optimization tactics on the outcomes.
of the parametric values for each crop based on the
training and data collection procedures previously REFERENCES:
outlined. The crop label can now be used by the [1]. N. H. Kulkarni, G. N. Srinivasan, B. M.
system to modify all of the numbers. The actuators Sagar and N. K. Cauvery, "Improving
will adjust these values and maintain the value of Crop Productivity Through A Crop
each parameter sufficiently stable to prevent Recommendation System Using
overflow or negligence after all sensor data has Ensembling Technique," 2018 3rd
been gathered, processed, and stored in the system. International Conference on
The technology will be able to calculate Computational Systems and Information
the quantity of humidity and temperature for each Technology for Sustainable Solutions
crop based on these facts. While adding more (CSITSS), 2018, pp. 114-119, doi:
values to other elements, the system is refined 10.1109/CSITSS.2018.8768790
enough to observe the current state and outcome of [2]. Vogel E, Donat MG, Alexander LV,
the crop. These data can be used by the system to Meinshausen M, Ray DK, Karoly D, et al.
make judgments. The amount of N, P, and K, The effects of climate extremes on global
among other parameters not shown in this graph, agricultural yields. Environmental
has also been graphed and data logged in a Research Letters. 2019; 14(5):054010.
comparable manner. The outcomes are displayed in https:// doi.org/10.1088/1748-
Figure 4, where we can observe that the highest 9326/ab154b
accuracy (DRL) of the random forest-based [3]. Z. Doshi, S. Nadkarni, R. Agrawal and N.
classifiers and the DRL-based classifiers are nearly Shah, "AgroConsultant: Intelligent Crop
equal. The rationale is that a deep neural network's Recommendation System Using Machine
overall performance grows with dataset length. Learning Algorithms," 2018 Fourth
Nonetheless, based on the existing dataset, DRL's International Conference on Computing
accuracy surpasses random forest's as the dataset Communication Control and Automation
lengthens. Next, the accuracy of different machine (ICCUBEA), 2018, pp. 1-6, doi:
learning systems are presented, including KNN, 10.1109/ICCUBEA.2018.8697349.
Naïve Bayes, and decision tree classifiers. [4]. Anshal Savla, Parul Dhawan, Himtanaya
Bhadada, Nivedita Israni, Alisha
VI. CONCLUSION AND FUTURE Mandholia, Sanya Bhardwaj (2015),
WORK: Survey of classification algorithms for
In this study, we investigated a number of formulating yield prediction accuracy in
agricultural data analysis methods offered by precision agriculture, Innovations in
various academics. With the least amount of input, Information, Embedded and
precision agriculture leverages technology to give Communication Systems (ICIIECS).
farmers the best potential results. The automation [5]. Aakunuri Manjula, Dr.G .Narsimha
of intelligent behavior, or artificial intelligence, is (2015), XCYPF: A Flexible and
continuously improving our society and helping Extensible Framework for Agricultural
humanity in a variety of ways. The impact of IoT Crop Yield Prediction, Conference on
and AI on creative farm management is examined, Intelligent Systems and Control (ISCO)
and a brief introduction is given to the most [6]. Yash Sanghvi, Harsh Gupta, Hamish
popular machine learning methods in precision Doshi, Divya Koli, Amogh Ansh Divya
agriculture. We used decision tree, random forest, Koli, Umang Gupta (2015), Comparison

DOI: 10.35629/5252-0602575580 |Impact Factorvalue 6.18| ISO 9001: 2008 Certified Journal Page 579
 International Journal of Advances in Engineering and Management (IJAEM)
Volume 6, Issue 02 Feb 2024, pp: 575-580 www.ijaem.net ISSN: 2395-5252

of Self Organizing Maps and Sammon's [13]. Kumar, A., Kumar, N., & Vats, V. (2018).
Mapping on agricultural datasets for Efficient crop yield prediction using
precision agriculture, International machine learning algorithms. International
Conference on Innovations in Information, Journal of Research in Engineering and
Embedded and Communication Systems Technology, 05(6), 3151–3159.
(ICIIECS). [14]. Renuka & Terdal, S. (2019). Evaluation of
[7]. R. Kumar, M. P. Singh, P. Kumar and J.P. Machine learning algorithms for Crop
Singh, "Crop Selection Method to Yield Prediction. International Journal of
maximize crop yield rate using machine Engineering and Advanced Technology,
learning technique," 2015 International 8(6).
Conference on Smart Technologies and [15]. Yu Ding, Liang Ma, Jian Ma, Mingliang
Management for Computing, Suo, Laifa Tao, Yujie Cheng, Chen Lu,
Communication, Controls, Energy and Intelligent fault diagnosis for rotating
Materials (ICSTM), 2015, pp. 138-145, machinery using deep Q-network based
doi: 10.1109/ICSTM.2015.72254034 health state classification: A deep
Smart Technologies and Management for reinforcement learning approach,
Computing, Communication, Controls, Advanced Engineering Informatics,
Energy, and Materials (ICSTM). Volume 42, 2019, 100977, ISSN 1474-
[8]. A. T. M. S. Ahamed et al., "Applying data 0346,
mining techniques to predict annual yield https://doi.org/10.1016/j.aei.2019.100977.
of major crops and recommend planting
different crops in different districts in
Bangladesh," 2015 IEEE/ACIS 16th
International Conference on Software
Engineering, Artificial Intelligence,
Networking and Parallel/Distributed
Computing (SNPD),2015, pp. 1-6,
doi:10.1109/SNPD.2015.7176185.
[9]. Tapas Ranjan Baitharua, Subhendu Kumar
Panic (2016), Analysis of Data Mining
Techniques for Healthcare Decision
Support System Using Liver Disorder
Dataset' International Conference on
Computational Modeling and Security
(C.M.S.).
[10]. Aymen E Khedr, Mona Kadry, Ghada
Walid (2015), Proposed Framework for
Implementing Data Mining Techniques to
Enhance Decisions in Agriculture Sector
Applied Case on Food Security
Information Center Ministry of
Agriculture, Egypt, International
Conference on Communications,
management, and Information technology
(ICCMIT').
[11]. Monali Paul, Santosh K. Vishwakarma,
Ashok Verma (2015), Analysis of Soil
Behaviour and Prediction of Crop Yield
using Data Mining Approach,
International Conference on
Computational Intelligence and
Communication Networks.
[12]. Shastry, A., Sanjay, H. A., & Bhanushree,
E. (2017). Prediction of crop yield using
Regression Technique. International
Journal of Computing, 12(2), 96–102.

DOI: 10.35629/5252-0602575580 |Impact Factorvalue 6.18| ISO 9001: 2008 Certified Journal Page 580