Smart Irrigation System using Machine Learning and IoT

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 06 | Jun 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 2108
Smart Irrigation System using Machine Learning and IoT
Bharath D.A.1, S. Amith Nadig2, Manjunath G.S.3
1VIII Semester, Dept. of ISE, BNMIT
2 VIII Semester, Dept. of ISE, BNMIT
3Asst. Professor, Dept. of ISE, BNMIT, Karnataka, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - To realise IoT promise in commercial-scale
applications, integrated InternetofThings(IoT)platforms are
required. The key challenge is to make the solution flexible
enough to fulfil the demands of specific applications. A IoT-
based platform for smart irrigation with a flexible design is
created so that it allows developers to quickly link IoT and
machine learning (ML) components to create application
solutions. The design allows for a variety of customised
analytical methods to precision irrigation, allowing for the
advancement of machine learning techniques. Impacts on
many stakeholders may be predicted, includingIoTspecialists,
who would benefit from easier system setup, andfarmers, who
will benefit from lower costs and safer crop yields.
The typical irrigation procedure necessitates a large quantity
of water use, which results in water waste. An intelligent
irrigation system is desperately needed to decrease water
waste during this tiresome process. Machine learning (ML)
and the Internet of Things (IoT) have made it possible to
develop an intelligent system that can accomplish this
operation automatically and with minimum human
intervention. An IoT-enabled ML-trained recommendation
system is suggested in this paper for optimum water
consumption with minimal farmer interaction. In the
agriculture field, IoT sensors are used to captureexactground
and environmental data. The collected data is transferredand
kept in a cloud-based server that uses machine learning to
evaluate the data and provide irrigation recommendations.
Key Words: IoT, ML, cloud, irrigation, water.
1. INTRODUCTION
In India, where agriculture accounts for 60-70percentofthe
GDP, there is a pressing need to modernise traditional
agricultural techniques in order to increase output. The
groundwater table is lowering day by day as a result of
uncontrolled water usage; lack of rainfall and shortage of
land water also contribute to a decrease in the amount of
water on the planet. Water scarcity is currently one of the
world's most pressing issues. Water is required in every
sector. Water is also necessary in our daily lives.
Agriculture is one of the industries that need a lot of water.
Water wastage is a serious issue in agriculture. Every time
there is a surplus of water, it is distributed to the fields.
Climate change and its consequences are widely explored in
academic studies on water resources and agriculture.
Because of the potential repercussions of global warming,
water adaptation methods are being considered to assure
water availability for food and human production as well as
ecosystem sustainability.Additionally,thesafetyofwater for
human consumption and return to the environmentmust be
maintained. Increased water shortages, poor quality of
water, higher water and soil salinity, loss of biodiversity,
increased irrigation needs, and the expense of emergency
and corrective action are all possible risks from climate
change. As a result of these factors, a rising number of
research are focusing on creating creative water utilisation
in irrigation. The Internet of Things (IoT) has now
progressed from a concept to being implemented in real-
world applications. Since then, the technological and
application hurdles have been considerable.
IoT platforms enable complex real-time control systems by
layering communication infrastructure,hardware,software,
analytical approaches, and application knowledge.
Recognizing the expected IoT consequences on systems is
one of the most difficult technological issues, because IoT
allows systems to becomeservicemashups,combiningitems
as services. System development will become dynamicplug-
and-play interoperable service composition, and system
logic will become service orchestration as a result.
An IoT-based smart irrigation system with an effective
machine learning algorithm is developed toassistfarmers in
overcoming the uncertainty of rainfall and increasing
production. This model provides a superior irrigation
decision-making model. This research presents a Machine
Learning (ML) strategy for successfully regulating irrigation
and enhancing agricultural yield as a result.
2. LITERATURE REVIEW
Goldstein et al. (2017) [1] suggested a recommendation-
based irrigation management system that combined
machine learning with agronomic knowledge. According to
the system, the best regression model with 93 percent
accuracy, and the best classifier model with 95 percent
accuracy, Gradient Boosted Regression Trees and Boosted
Tree Classifier, provide superior irrigation prediction
decisions than the linear regression model. To assist the
agronomist in making better selections, the models were
trained with eight separate sets of features. The Internet of

Multimedia Things (IoMT) was proposed by Al Zu'bi etal.[2]
(2019), which focuses on the use of multimedia sensors in
the field for irrigation optimization.Digital image processing
is used to monitor crops and soil.
The multimedia sensor sends the collected photos of the
crops to the image processing system, which makes the
choice based on the proportion of soil cracks.Thismakesthe
Future No-Man irrigation management system possible.
According to the data mining technique, Rushika Ghadge et
al. (2018)[3] created a system that employs bothsupervised
and unsupervised ML algorithms to forecast crop and soil
quality and kind of land, as well as assess the nutrients
present in the soil to boost agricultural yield. This effort
assists farmers in cultivating healthier crops in the proper
soil to increase yield, as well as serving as a conduit for
providing timely information to farmers regarding crop
quality and nutrient requirements. For soil moisture
estimation, a learning model based on Support Vector
Regression (SVR) and K-means clustering was created.
Humidity, radiation, soil moisture, air, and soil temperature
were all captured in the field and sent into the training
system of SVR model.
To increase accuracy and reduce error rate, the SVR model
output is sent to K-means clustering. For optimal yield
management, the final output from k-means is employed to
govern the water pump controller. However, the majority of
the older prediction models had a large variance, which
causes the machine learning model to perform poorly.
Ensemble learning, as described by Zhao et al. (2018)[4],
Catolino and Ferrucci[5] (2018), Joshi and Srivastava[6]
(2014), and Ren et al.[7] (2016), may be utilised to deal with
such significant variation. By combining different learning
models to predict the output of a single system, ensemble
approaches improve performance.
Some ensemble approaches, particularlybagging,have been
shown to decrease the problem of overfitting and
underfitting of training data. The Bootstrap Aggregation
approach improves single regression trees by using many
models, each of which is trained using randomly selected
samples from the original dataset.The baggingapproachhas
a smaller prediction error than the other single models.
Gonzalez et al. (2014)[8] provide a bagging method for
forecasting power price that is compared to the random
forest approach for both classification and regression
models.
A complete literature study was conducted, and the paper
suggests some of the most efficient feasibletechnologiesand
algorithms for the creation of a Smart Farm Monitoring
System based on the findings of the literature research and
experiments. Ersin et al.[9] suggested a microcontroller-
based irrigation system that is more efficient and cost-
effective than other traditional techniques. Liu et al.
described precision irrigation technologies. [10].Agrawal et
al. presented a smart irrigation system using Raspberry Pi
and Arduino. [11]. Koprda et al.presenteda microcontroller-
based irrigation solution. Ahouandjinou et al. discuss farm
pest detection using ultrasonic sensors in their paper [12].
[13]. Goap et al.[14] provided a full overall design foranIoT-
based irrigation system.
Smith et al.[15] presented machine learning algorithms for
soil categorization. Wu et al.[16] investigated a farm vehicle
and smart dispatching strategy. Ryu et al. [17] presented an
integrated method to smart farming. Kwok et al. [18]
proposed utilising deep learning to recognise plants and
then determiningtheoptimal wateringvolumedependingon
plant type. Wang, Muzzammel, Raheel, and colleagues[19]
explored deep learning and an altitude-based economical
irrigation technique. A WSN technique for precisionfarming
was presented by Martinell et al. [20]. Izquierdo et al. [21]
suggested a smart farming solution based on cloud andedge
computing.
Bacco [22] provided a detailed smart farming method in
their study effort, including all constraints, enablers, and
prospects. This paper emphasises and presents a
comprehensive, precise picture of the feasible answer for
agricultural demands following a thorough analysis of
currently accessible literature that dealswithcontemporary
farming challenges and their associatedsolutions.The paper
describes a distributed sensor network field whose
prototype was created for this research.
3. METHODOLOGY
3.1. Existing Methods
In general, there is no automated irrigation method that is
being used all over the world. However,somestudyhasbeen
done in the topic of automating the watering process. In
most existing studies, the following is the basic method for
Automated Irrigation: To begin, data is collected from
several sensors to determine the moisturecontentofthesoil
and the temperature of the surrounding environment.
They're attached to a breadboard that's wired up to the
Arduino board. The Arduino IDE receives the data from the
board. The programming language employed executes
instructions that extract data and reflect it, i.e., a decision is
made whether to turn the water pump "On" or "Not" based
on the extracted data.
3.2. Proposed Solution
Step 1: As demonstrated in Fig. 1, irrigation can be
automated utilising sensors, microcontrollers,Wifimodules,
and the ThingSpeak platform. A controller is necessary to
maintain all of the sensors and to drive the motor as needed.
We utilised NodeMcu to accomplish this. The NodeMcu can
output a maximum voltage of 5 volts. The moisture sensors
module and DTH11 sensor can both be powered by 5 volts,
but not the motor. We need at least 7 volts to run a motor. To

solve this issue, we utilised a 9v battery to power the motor.
We'll need a switch to regulate the motor whenever it's
needed. We utilised a relay module to accomplish this. It's a
switch in the electrical system. We must deliver a strong
pulse to the module in order to close the switch. The field is
constantly monitored by the soil moisturesensor.NodeMCU
is attached to the sensors. The sensor data is sent to the user
via wireless transmission so that he can manage irrigation.
Step2: Finally, all the data needs to be there in ThingSpeak
for visualisation. by using write API keys we will send the
sensor data to the server. In ThingSpeak wecanvisualisethe
data of every sensor over the time.
Step3: Fetching the data from ThingSpeak to our python
script
1. Import all required libraries (json, urllib.request).
2. Create anAPIusingREAD_API_KEYandCHANNEL_IDof
ThingSpeak.
3. Request the ThingSpeak websitebyusingurllib.request
module.
4. Store the json response from ThingSpeak.
5. Retrieve temperature, Humidity, Soil moisture values
from json data.
Step4:
1. The weather data was obtained through the Kaggle
platform.
2. The performance of rainfall prediction is benchmarked
using a variety of learning methods in this work. These are
the supervised learning methods NB, C4.5, SVM, ANN, and
RF.
3. An ensemble of the above models is used to train a Voting
Classifier, which predicts an output (class) based on the
output's highest likelihood. It simply sums up the results of
each classifier fed into the Voting Classifier and predicts the
output class with the most votes. We propose a single model
that trains on numerous models and predicts output based
on the cumulative majority of votes for each output class,
rather than building separate specialised models and
determining their performance.
Step5: Taking the final decision
When the soil moisture falls below a certain threshold, a
motor will switch on. Instead of turning on the motor
immediately, we examine the probability of rain using the
above-mentioned ensemble methodologies, and if rain is
likely to occur during that time period,wewaita while.From
crop to crop, the threshold level will differ. If a crop need
more water, we will increase the high threshold level sothat
the crop receives more water. Alternatively, if the crop
requires less water, we will specify a low value
4. IMPLEMENTATION
Hardware used:
1. NodeMcu ESP8266
2. Soil Moisture Sensor Module
3. Submersible DC motor
4. DTH11 sensor
5. Relay module
Software used:
1. Arduino IDE
2. ThingSpeak
3. Google Sites
NodeMcuEsp8266: It's an IoT gadget that's open-source. It's
a 32-bit microcontrollerthatallowsWi-Fi-connectedgadgets
to send and receive data. It's a low-cost semiconductor with
TCP/IP networking software built in. Thereare17GPIOpins
on this board. It contains a Tensilica L 106 RISC CPU that
uses very little electricity. It's compatible with ADCs, power
amplifiers, and certain power management modules are all
available. It contains 4KB of memory storage. Figure 1
depicts NodeMcu in its most basic form.
Fig - 1:NodeMcu
Soil Moisture Sensor Module: The purpose of a soil moisture
sensor is to determine the amount of water in the soil. It
comprises mostly of a pair of conducting probes.Thechange
in resistance between those probes is used to calculate the
moisture content. The quantity of moisture in the soil hasan

inverse relationship with resistance. It transmits analogue
data.
The value will vary from 0 to 1023 after feeding this into
ADC. As a result, if there is no water in the soil, the value
decreases. 1023 will be the number. So for changing this
value into percent we need map (0,1023) to (1,100) which
can be done using map function.
Fig - 2: Soil Moisture Sensor.
Submersible DC: motor is one that can be completely
immersed in water. In order to prevent water from entering
the motor, it is hermetically sealed. It converts rotational
energy into kinetic energy, which is then converted into
pressure energy, which pushes water to the surface. This
engine will be submerged in water, with a conduit
connecting it to the water's output.
Fig - 3: Submersible pump
DTH11 sensor: It's a multi-purpose sensor that measures
temperature and humidity in the environment. It is made up
of humidity detecting material and a temperature sensing
thermistor. A humidity detecting material isa capacitorwith
humidity as a dielectric substancebetweenthem,causing the
capacitance to alter as the humiditychanges.Weunderstand
how thermistors function. The resistance value fluctuatesas
a function of temperature. It operates at 3-5 volts, which we
can acquire from the NodeMcu.
Fig - 4: DTH11 Sensor
Relay Module: It's a type of electrical switch that works by
using magnetism. The Relay module's primary function is to
control the motor. The NodeMcu's maximum output voltage
is 5 volts, which is insufficient to drive the motor.So,todrive
the motor, we'll connect the relay module to the NodeMcu,
and power the module with a 9v battery. We will send a high
to low pulse to the relay module anytime we want to turn on
the motor, the switch will shut, and 9v will be sent to the
motor. The single channel Relay module is shown in Fig 5 as
an example
Fig - 5: Relay Module
5. ALGORITHMS
The categorization technique C4.5 is one of the most
effective. C4.5 generates a decision tree, where each node
divides the classes according to the information. The
property with the highest normalised information gain is
used to determine the splitting criteria. Humidity and
temperature, for example,areincludedinourdata collection.
The C4.5 algorithm investigates these aspects first to
determine which is optimal for data splitting (a feature with
maximum information gain). Afterthat,thefeatureisutilised
to partition the dataset into the following feature till it
reaches the final destination. The algorithm's output is
shown in Table.
Naïve Bayes
Nave Bayes is a supervised machine learning model that
belongs to the family of probabilistic classifiers that applies
the Nave Bayes theory to the dataset's independence
assumption. By computing the assumptions, Nave Bayes
determines the probability of each feature in the dataset.
Nave Bayes determines every attribute conditional
probability on the class label for every known class label.
The product rule is then used to calculate the joint
conditional probability for the characteristics of a label. The
Nave Bayes model is then used to derive the conditional
probability for the class characteristics. The class with the
highest probability is provided afterperformingthismethod
for each class value. The results of the algorithm are shown
in Table.
Support Vector Machine
The support vector machine is a machine learning model
that uses a hyperplane to partition a dataset into two pieces.
This partitioning procedure treats each class label

separately, and it may be done by classifying the data into
class A and not class B, where A and B are the two class
labels. Calculating the Euclidean distance between eachdata
point and the hyperplane's margin is used to classify the
data. When data cannot be linearly separatedina lowerlevel
space, the Support Vector Machine model employs a kernel,
which is a set of scientific functions, to allow for data
categorization in a complicated dimensional space. In
machine learning, many kernel functions, such as radial, are
available to regulate the above.
Neural Networks
Exhibit machines that mimicked the brain's functions
influenced the development of neural networks.Everybrain
unit is linked to a slew of others. In terms of the initial state
effect of the linked neuronal units, links might be either
enforcing or inhibitory.Asummingfunctionmightbeusedto
unite the input values of each individual brain unit. This
model is utilised in regression and classification, as well as
prediction and clustering.Therearetwoprimaryfactorsthat
have a substantial impact on the performance of neural
network classifiers. The number of hidden layers is the first,
and the value of the learning rate is the second. The results
of the algorithm are shown in Table IV.
Random Forest
Random forest is a machine learningmodel thatmaybeused
for prediction, regression, and classification, among other
things. This algorithm is an ensembleofdecisiontree models
that aims to produce a multiplicity of decision tree models
from the same training data and generate the final class as
the output. The number of characteristics to freely study
(Num Features), maximum depth of the tree (Max Depth),
and number of trees (Num Tree) parameters aremodified in
the random forest classifier. The findings of the research
show shows the Random Forest classifier's classification
performance improves as the number of features, trees, and
depth grow. The results of the algorithm are shown inTable.
6. RESULTS AND DISCUSSIONS
A. C4.5 Algorithm
Accuracy: 0.77
B. Naïve Bayes
Accuracy: 0.81
C. Support Vector Machine
Accuracy: 0.82
D. Neural Networks
Accuracy: 0.77
E. Random Forest
Accuracy: 0.75
Machine learning-based prediction performance varies
amongst algorithms, with the artificial neural network
approach having a modest performance edge over other
categorization models. Each model has certain flaws,butthe
overall outcome is always better since the error rate is
reduced. When one model fails, other models will step in to
help. Because we're employing ensemblelearning,we'll take
into account a mix of model knowledge. a majority of people
A voting model is constructed that trains on a mixture of the
aforementioned models and predictsanoutput(class)based
on the majority of the highest likelihood of each model's
preferred class as the output.Itsimplyaggregatestheresults
of each machine learningmodel fedintoVotingClassifier and
predicts the output class based on the most votes.
The representation of detected data via sensors in
ThingSpeak is shown in Figures 11,12,13. It aids in the

interpretation of data. We may use this data to combine,
transform, and create new data, and we can use built-in
charting algorithms to graphically grasp the relationships
between the data. In the future, we'll beabletocombinedata
from numerous sources to provide a more complex study.
Fig - 6: Field Chart 1
7. CONCLUSION AND FUTURE ENHANCEMENT
Regular crop updates, such as moisture, humidity, and
temperature, are critical in agriculture. Climate forecasting
data accuracy has increased dramatically as a result of
technological advancements, and weather forecasting data
may now be utilised to estimate rainfall ina specificlocation.
To estimate rainfall possibilities, this study suggests an
Automated Irrigation System thatusestheInternetofThings
and Ensemble Learningtechniques.Thesuggestedtechnique
predicts rainfall in the near future by combining sensordata
from the recent past with weather projected data. We
utilised the Ensemble learning approach to forecast the
likelihood of rain on that particular day. Rather than
constructing separate specialised models and calculating
classification metrices for each of them, the main purpose of
this technique is to develop a single model that trains
numerous models and classifies the output based on their
aggregate majority of votes for each outputclass.Forecasted
rainfall possibilities are superior in terms of accuracy and
mistake rate. A solo system prototype can also use the
prediction method. The system prototype is low-cost
because it is based on open-source technologies. We'dliketo
perform a water-saving study based on the suggested
technique in the future, with more nodes anda lowersystem
cost. The irrigation system automation we provided as part
of our strategy performed wonderfully. It's also cost-
effective. Using this technique, we can reduce the number of
people needed in the fields for upkeep. This approach will
not only irrigate the ground automatically based on the
moisture level in the soil and the possibility of rain, but it
will also send the data to the Thingspeak server,allowingthe
farmer to keep track of the land's status.
REFERENCES
[1]. Muhammad Ayaz, Mohammad Ammad-Uddin, Zubair
Sharif, Ali Mansour And El-Hadi M. Aggoune (2019).
“Internet-Of-Things (Iot)-Based Smart Agriculture:
Toward Making The Fields Talk”. SpecialSectionOnNew
Technologies For Smart Farming 4.0.
[2]. HamzaBenyezz, MounirBouhedda,SamiaRebouh(2021).
“Zoning irrigation smart system based on fuzzy control
technology and IoT forwater andenergysaving”.Journal
of Cleaner Production Volume 302, 15 June 2021,
127001
[3]. Vaishali S, Suraj S, Vignesh G,Dhivya SandUdhayakumar
S (2017). “Mobile Integrated Smart Irrigation
Management and Monitoring System Using IOT”.
International Conference on Communication and Signal
Processing, April 6-8, 2017, India.
[4]. NurzamanAhmed,DebashisDe,andMd.IftekharHussain
(2018). “Internet of Things (IoT) for Smart Precision
Agriculture and Farming in Rural Areas”. IEEE

INTERNET OF THINGS JOURNAL, VOL. 5, NO. 6,
DECEMBER 2018
[5]. Jhonn Pablo Rodríguez, Ana IsabelMontoya-Munoz,
Carlos Rodriguez-Pabon, Javier Hoyos, Juan Carlos
Corrales (2021). “IoT- Agro: A smart farming system to
Colombian coffee farms”. Computers and Electronics in
Agriculture Volume 190, November 2021, 106442
[6]. Md. Rezwan Hossain Naeem, Shadman Gawhar, Md.
Belawal Hoque Adib, Sanjid Ahmed Sakib, Abir Ahmed
and Nafiz Ahmed Chisty(2021). “An IoT Based Smart
Irrigation System”. 2021 2nd International Conference
on Robotics, Electricaland Signal ProcessingTechniques
(ICREST)
[7]. H.G.C.R.Laksiri, H.A.C.Dharmagunawardhana,
J.V.Wijayakulasooriya “Design and Optimization of loT
Based Smart Irrigation System in Sri Lanka”. 2019 IEEE
14th International Conference on Industrial and
Information Systems (ICIIS), 18-20 Dec.,Peradeniya, Sri
Lanka
[8]. Neha K.Nawandar, Vishal R.Satpute(2019). "IoT based
low cost and intelligent module for smart irrigation
system”. Computers and Electronics in Agriculture
Volume 162, July 2019, Pages 979-990
[9]. A. Anitha, Nithya Sampath, M.Asha Jerlin(2020). “Smart
Irrigation system using Internet of Things”. 2020
International Conference on Emerging Trends in
InformationTechnologyandEngineering(ic-ETITE)978-
1-7281-4142-8/20/$31.00 ©2020 IEEE 10.1109/ic-
ETITE47903.2020.271
[10]. Thilina N. Balasooriya, Pranav Mantri, Piyumika
Suriyampola (2020). “IoT-BasedSmartWateringSystem
Towards Improving the Efficiency of Agricultural
Irrigation”. 2020 IEEE Global Conference on Artificial
Intelligence and Internet of Things (GCAIoT) | 978-1-
7281-8420-3/20/$31.00 ©2020 IEEE | DOI: 10.1109 /
GCAIOT51063.2020.9345902

Smart Irrigation System using Machine Learning and IoT

Recommended

More Related Content

Similar to Smart Irrigation System using Machine Learning and IoT (20)

More from IRJET Journal (20)

Recently uploaded (20)

Smart Irrigation System using Machine Learning and IoT