Towards a Novel Approach for Smart Agriculture Predictability

Rima Grati

1 a

, Myriam Aloulou

2 b

and Khouloud Boukadi

3 c

Zayed University, College of Technological Innovation, Abu Dhabi, U.A.E.

Liwa College of Technology, Abu Dhabi, U.A.E.

Miracl Laboratory, Faculty of Economics and Management, University of Sfax, Tunisia

Keywords:

IoT, Smart Agriculture, Machine Learning, Self Organization Map, Machine Learning, Deep Learning.

Abstract:

The practice of growing crops and raising cattle is the traditional method of agriculture, a primary source of

livelihood. The introduction of advanced technologies and tools provides solutions to predict and avoid soil

erosion, over-irrigation, and bacterial infection for crops. Machine learning and Deep learning solutions are

hitting high results in terms of precise farming. The most challenging factors for research society are iden-

tifying the water need, analyzing soil conditions and suggesting the best crops to cultivate, and predicting

fertilizer amounts to prevent bacteria. Grouping similar features helps with accurate prediction and classiﬁ-

cation. Considering this, we introduce an integrated model Group Organize Forecast (GOF), using Machine

Learning (ML) and Deep learning (DL) techniques to balance the requirements and improve automatic irri-

gation. GOF analyzes the irrigation requirement of a ﬁeld using the sensed ground parameters such as soil

moisture, temperature, weather forecast, radiation levels, the humidity of the crop ﬁeld, and other environmen-

tal conditions. We use a real-time unsupervised dataset to analyze and test the model. GOP clusters the data

using Self Organizing Map (SOM) organizes the classes using Cascading Forward Back Propagation (CFBP),

and ﬁnally predicts the requirement for water and solution to control bacteria in the near future.

1 INTRODUCTION

The Internet of Things (IoT) is an integrated tool used

to sense data from multiple devices and evaluate the

internal and external state of communication. IoT ap-

plications in smart agriculture focus on crop water

management, pest control, temperature adjustments,

precise detection, and nutrient management with safe

storage methods. According to a study, (Baktha-

vatchalam et al., 2022) 64% of cultivation depends

on the monsoons, during which irrigation needs 85%

of the water, of which 60% is wasted in the process.

Analyzing the features to monitor the water-level and

temperature, which prevent the bacteria from growing

based on the changes in the environment, is a major

concern of the research community.

Machine learning is a popular technology and a

branch of artiﬁcial intelligence that allows comput-

ers to learn without explicit programming. Machine

learning techniques are used to extract the features

and group the required input into clusters. Deep

https://orcid.org/0000-0002-6995-465X

https://orcid.org/0000-0001-8404-7487

https://orcid.org/0000-0002-6744-711X

learning techniques are suitable to analyze, classify,

and predict the requirements for sustainable irrigation

management. Both techniques are effective decision-

support tools for precision agriculture. Traditional

framing includes manual decisions on controlled wa-

ter management, crop selection, weather forecasting,

and analyzing soil conditions. This can be enhanced

and improved based on the needs of the crops using

an integrated techniques (Pathan et al., 2020).

1.1 Motivation and Contribution

The lack of proper measures for taking the right de-

cision in smart irrigation is a big challenge for the

smart agriculture industry (Ben Abdallah et al., 2023).

Prediction of the suitable crop yield, analysis of the

water requirement, and equal distribution of soil nu-

trition are the major requirements to be considered

in the development of an intelligent IoT irrigation

model. Available research models are designed for a

speciﬁc ﬁeld structure and depend on the sensor fea-

tures (Ayaz et al., 2019). (Sarker, 2021) highlighted

some of the deep learning techniques and their impor-

tance to handling the challenges faced by IoT agricul-

Grati, R., Aloulou, M. and Boukadi, K.

Towards a Novel Approach for Smart Agriculture Predictability.

DOI: 10.5220/0012082400003538

In Proceedings of the 18th International Conference on Software Technologies (ICSOFT 2023), pages 96-105

ISBN: 978-989-758-665-1; ISSN: 2184-2833

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

ture. It is highly recommended to develop a general-

ized model for predicting the requirements for water,

food, air, temperature, and other necessities for the

plant. We show a novel prediction model to protect

and develop an IoT agriculture sector with the follow-

ing contributions.

• Feature extraction: GOF analyzes the required

features and clusters them into similar groups

using the unsupervised clustering technique Self

Optimization Map (SOM). This helps to predict

water requirements, the presence of bacteria, and

the suitable crop.

• Classiﬁcation and prediction model: GOF catego-

rizes the classes using Cascading Forward Back

Propagation Neural Network (CFBPNN) model.

CFBPNN identiﬁes the pattern of the variable and

the requirements for irrigation.

• Decision Tree and correlation Analysis: GOF de-

cides the prediction factors using the decision tree

algorithm and visualizes them using the correla-

tion plot technique. The optimal cluster helps in

prediction using decision rules. Finally, SOM pre-

dicts the water requirement and the factors inﬂu-

encing the spread of bacteria.

1.2 Organization of the Paper

The rest of the paper is organized as follows. Sec-

tion 2 reviews some machine learning and deep

learning-based models and their techniques for the

smart irrigation system. Section 3 Focus on the SOM

clustering method to divide similar groups and clas-

sify the target data using the CFBPNN classiﬁcation

model. An evaluation metric with state-of-the-art

comparison with GOF is given in Section 4. Finally,

a logical conclusion is drawn in Section 5.

2 RELATED WORK

In agriculture, Machine Learning (ML) is used to

forecast soil parameters like organic carbon and mois-

ture content; it also predicts diseases and weeds in

crops and identiﬁes species. Remote monitoring of

ambient and soil characteristics is used for agronomic

applications to predict crop health. A sensor-based

network is being used to forecast the watering sched-

ule for agricultural ﬁelds. A wireless sensor network

collects data from external variables such as pressure,

humidity, temperature, soil moisture, salinity, and

conductivity. Agricultural applications can be made

incredibly simple and efﬁcient using three stages of

machine learning as data acquisition, model develop-

ment, and generalization. Deep Learning (DL) meth-

ods over traditional machine learning is enhanced by

adding additional complexity to the model and change

the input with a range of functions that hierarchi-

cally allow data representation, based on the network

architecture, through multiple levels of abstraction

(Jimenez et al., 2021). Selecting appropriate input

variables as temperature, humidity, moisture of soil,

and wind ratio helps identify the water requirement

for the plant. DL techniques are used to trace the op-

timistic features and predict the requirements.

Compared to ML, DL models are popular in es-

timation of image and sound processing. Some of

the popular DL models used for optimizing irriga-

tion decision are Artiﬁcial Neural Network (ANN),

Recurrent Neural Network (RNN) with loop connec-

tivity, Long Short-Term Memory (LSTM), and Con-

volutional Neural Network (CNN). Precision irriga-

tion is the solution to deliver bigger, better, and more

proﬁtable yields with fewer resources. Some of the

applications using the above methods are to identify

the best crop, detect factors that destroy the crops, an-

alyze the presence of diseases to obtain insights about

crop growth and help in decision making. Some of

the smart agricultural frameworks proposed in recent

years are given in Table 1.

The main aim of the study to perform unsuper-

vised clustering method, to group set of variables and

identify the classes having impact over the change.

Some of the research models using cluster techniques

are discussed in what follows:

Application of Clustering Smart Agriculture.

Clustering is aimed to identify a distinct group, based

on the similarity of a given dataset, while the arrange-

ment of data into clusters results in low inter-cluster

similarity and high inter-cluster similarity (Swamy-

nathan, 2019). An integrated study using fuzzy time

series techniques and clustering to manage wireless

sensor nodes plotted on the agricultural ﬁeld is pro-

posed by (Prabhu et al., 2014). The study concluded

that the proposed model improved the energy efﬁ-

ciency of the sensors with real-time monitoring of the

farm. It has been observed that the low yields are

caused due to attacks from pests resulting in inade-

quate irrigation. This has been investigated using the

K-Means clustering technique with image data by en-

abling the Wireless Sensor Network (WSN) for smart

irrigation by (Nisha and Megala, 2014). The image

data of plant leaves are segmented into cluster groups

based on the feature similarities, the model gave im-

proved results compared to other WSN-based irriga-

tion systems with clustering models for pest detec-

Towards a Novel Approach for Smart Agriculture Predictability

Table 1: Smart agriculture models using ML and DL techniques.

Reference Model Results

(Mehra et al., 2018) Deep Neural Network

(DNN), Artiﬁcial Neural

Network (ANN)

Classify and Control actions.

(Varghese and Sharma, 2018) Support Vector Ma-

chine(SVM)

Reduce manual labour, regular alerts

for complete ﬁeld control.

(Goap et al., 2018) Support Vector Regression

(SVR) and K-means

Soil Moisture Differences (SMD)

(Lavanya et al., 2020) Fuzzy logic Detect deﬁciency of nutrients from

the sensed data

(Rezk et al., 2021) wrapper feature selection,

and PART classiﬁcation

technique

suitable for crops: Bajra, Soybean,

Jowar, and Sugarcane.

(Reddy et al., 2020) Decision Tree(DT) algorithm Mail alert based on DT results re-

garding water supply in advance.

(Kashyap et al., 2021) Long Short-Term Memory

network (LSTM)

Predict the volumetric soil moisture,

and spatial distribution of water re-

quired to feed

(Akhter and Soﬁ, 2022) ML Linear regression model Categorised as safe an unsafe area

based on temperature and wetting du-

ration.

(Bakthavatchalam et al.,

2022)

MLP, Decision table,JRip Accurate prediction and implementa-

tion of precision agriculture

(Geetha Lekshmy et al.,

2022)

LSTM and random forest Water dispensed, pest detection with

images of ﬁeld object detection tech-

nique to avoid pests and animals.

tion. Other research by Ohana-Levi et al. (Ohana-

Levi et al., 2021) uses fuzzy K-means clustering, us-

ing the hierarchical method to identify the litigation

management zones in a citrus ﬁeld, to determine in-

ﬁeld variation and adjust site-speciﬁc irrigation man-

agement. Variables like: crop water stress, Normal-

ized Difference Vegetation Index (NDVI), digital sur-

face model, slope, aspect, and elevation were used for

the experiment. The study concludes that the inﬁeld

spatial variability is not constant among the variables

and within the orchard.

3 METHODOLOGY

The proposed methodology includes four stages given

in Figure 1. After collecting the live data from the

sensors, each feature is represented as a separate ele-

ment in the module /controller. The input features are

traced for a period and stored in a CSV ﬁle in the ﬁrst

stage. Secondly as part of the pre-processing tech-

nique, data normalization maintain the balance in the

data values, and the missing information is eliminated

to avoid misleads in the experiment. Finally the raw

data is compressed and divided into clusters using the

SOM Neural Network technique. And ﬁnally, the in-

put and class variable(target) are trained and tested

on the proposed CFBPNN prediction technique. The

training is repeated till the expected results are evalu-

ated.

3.1 Data Collection

In the PRECIMED project (PRIMA, 2023), we col-

lected real-time data provided by SENTEK technolo-

gies using the TEROS 12 soil Moisture and Electrical

Conductivity (EC) and Temperature Sensor. These

tools help collect Volumetric Water Content (VWC)

and monitor electrical conductivity and other re-

sources, such as temperature and the toxicity levels of

soil substances. VWC works with frequency-domain

technology, connected with sensors on 70MHz fre-

quency, to minimize salinity and textural effects.

TEROS 12 is the primary tool for measuring temper-

ature and electrical conductivity with a stainless-steel

electrode array and is accurate in mineral soils.

Data provided by the above sensors are:

• Volumetric Water Content (VWC) measurement

• Soil/substrate water balance

• Irrigation management

ICSOFT 2023 - 18th International Conference on Software Technologies

Figure 1: Procedure for proposed GOF integrated model.

• Soil Electrical Conductivity (EC) measurement

• Soil/substrate temperature measurement

• Solute/fertilizer movement

The live data is categorized with four controllers, each

controller supports with input features acquired from

various sensing devices.

The study gather the input from Finca experimen-

tal CEBAS-CSIC Santomera, (referred as controller

2). This provides 18 input features given in Table 2

with 1125 records collected from 28th February to 7th

March 2022 for training and 9th March 2022 to 15th

march 2022 for testing.

3.2 Data Pre-Processing

As the data collected from the sensors are not classi-

ﬁed under any of the categories: to develop a predic-

tion model, the classiﬁcation of variables under cer-

tain classes is mandatory. The raw data collected from

the sensors are not labeled, we are using unsuper-

vised learning methods as pre-processing technique,

to cluster the data and generate a target groups. Such

groups are considered clusters that help in analyzing

the relation between variables and help in tracing the

pattern of behavior.

Table 2: Input features provided by Controller 2.

Feature Details

Nivel de Bac-

teria

level of Bacteria present

t-uC Electric supply

Radiation

Solar

Energy received from solar panel

Precipitation Possibility of Rainfall

Rayos

Direction

Viento

wind direction

Temperature Humid and temperature in ratio

Precision -

Vapor

level of water vapour

xOrientation Direction to X axis

yOrientation Direction to Y axis

s1-0-

counts4WVC

sensor 1 Water Volumetric content

(WVC)

s1-0-temp Sensor 1 temperature ratio

s1-0-ec Sensor 1 electrical conductivity

(EC) measurement

s2-0-

counts4WVC

sensor 2 Water Volumetric content

s2-0-temp Sensor 2 temperature count

s2-0-ec Sensor 2 electrical conductivity

(EC) measurement

Humid Rela-

tive

Ratio of relative Humidity

HST Value of HST

Towards a Novel Approach for Smart Agriculture Predictability

3.2.1 Clustering

As part of dimensional reduction, and conversion of

data from unsupervised learning to supervised learn-

ing method we are using a neural network-based clus-

tering technique. Self Organizing Map (SOM), a deep

learning feature reduction technique introduced by

T. Kohonen (Kohonen, 1990). The SOM training

phase inputs the elements and the mapping is used

to classify the new input vector. A trained map clas-

siﬁes a vector from the input space with the closest

weighted node. In the process of training ”Euclidean

Distance” is computed for all weight vectors and com-

pared with other neurons in the space. The neurons

with similar input are considered the Best Matching

Unit (BMU) representing a lighter color in a visual

display. Other neurons closer to this are adjusted with

similar weights in the input vector.

1. Self Organizing Map (SOM) is used as a pre-

prossessing step for supervised learning.

2. SOM represents clusters by grouping similar data.

This reduces data dimensions and displays simi-

larities among data.

3. The reduction of dimensionality and grid cluster-

ing makes it easy to observe similarities in the

data for prediction.

4. SOMs factor in all the data in the input to gen-

erate these clusters and can be altered such that

certain pieces of data have more/less of an effect

on where input is placed.

5. Unlike other learning techniques in neural net-

works, training a SOM requires no target vector.

A SOM learns to classify the training data without

any external supervision.

Use of Clustering in the Study. In our study to pre-

dict the water requirement, variables like humidity,

wind rate, VWC, and temperature form a cluster. And

features like Bacteria level, precipitation of rainfall,

temperature, wind direction, humidity, water control,

and soil electrical conductivity (EC) as another clus-

ter.

The above-mentioned group of features has an im-

pact on predicting the water level or crop suitable for

the land. This is an excellent indicator to trace the nu-

trient availability and loss, soil texture, and available

water capacity. As the selected dataset does not have

any classiﬁcation variable, we have planned to imple-

ment the clustering technique to create multiple clus-

ters which classify the variables into related groups

using the SOM technique.

This helps in tracing the features which are closely

related and identifying the dependency for tracing the

target class.

SOM identiﬁes a winning neuron i

and updates

the weights of all other neurons with a certain distance

(d) using the Kohonen rule.

To implement the SOM we need to ﬁnd out the

best suitable cluster point, calculated from sums of

points to central distance. We have implemented the

K-means elbow point technique and pointed out that

x:4, indicating four clusters are suitable for the se-

lected dataset.

We have selected HEX-TOP as the topology func-

tion and LINK-DIST for the calculation of distance

between the variables with 0.9 ordering phase learn-

ing rate and 1000 as ordering phase steps with 0.02

tuning phase learning rate and 1.0 neighborhood dis-

tance and (2X2) map dimension and with random

weight and bias initialization. The model took 4 : 45

minutes to train the network, The SOM input planes

for four clusters and 18 input features are displayed in

Figure 2. The SOM technique divides the data into

Figure 2: SOM input planes with cluster available.

cluster data points based on similarity. The dataset is

represented by a special point, constructed by a learn-

ing method in the process of iteration. These points

are mapped on a 2D grid to represent a topology of

a honeycomb. The four clusters formed with similar

points for our study are given below:

• cluster 1 -214

• cluster 2-113

• cluster 3-626

• cluster 4-171

This method indicates the movement of the neighbor-

hood around the special point. Then these are ar-

ranged to form a cluster, of cells that are associated

with the same macroscopic phenomenon. From Fig-

ure 2 variables having similar cluster points fall un-

der the category of 2,4 clusters represented with black

shade. A list of variable that fall under cluster 2,4 are

given below:

• Input 4 - Precipitation of rain

• input 9 -x Orientation (wind movement)

• input 12 - s10temperature (sensor 1)

• input 15 - s20temperature (sensor 2)

ICSOFT 2023 - 18th International Conference on Software Technologies

100

From the above cluster grouping, we have grouped

variables that have close association: the wind move-

ment and the temperatures of both the sensors are

clusters to show the dependency on precipitation vari-

able. This cluster is used to represent the rainfall

based on features 9,12,15 given above.

3.3 Classiﬁcation

SOM trained the sample data and divided four sim-

ilar groups. As the next step classiﬁcation us-

ing Cascade Forward Back Propagation Neural Net-

work(CFBPNN) a time series-based technique is

trained and tested. The training and testing phase

helps in tracing a behavioral pattern and analyzing

the variation and then classifying the results. For ex-

ample, as discussed in the above section 3.2.1, each

cluster represents a target class for the prediction of

rainfall, water requirement, or bacteria identiﬁcation.

Examining the data and tracing the results based

on time and checking if it is categorized under the

same class is implemented using the CFBPNN classi-

ﬁcation technique.

Cascade Forward Back Propagation Neural Net-

work (CFBPNN) is the combination of feed-forward

with recurrent model. The proposed CFBPNN model

begins with a single input and adds multiple con-

nected layers, one by one in the process. Perceptions

are added one by one in this correlation; it starts with

a small number and ends up with a bigger size. Ad-

ditional connections improve the speed and learning

rate. We show the mathematical expression of the CF-

BPNN network in Equation 1.

y =

∑

i=1

+ f

∑

j=1

∑

i=1

(1)

In Equation 1, y represents the output layer,

∑

i=1

is used to calculate the sum of weights and bias of

each layer. The special feature of this network is to

carry forward the calculated weights and bias by es-

tablishing a direct relationship between the input and

hidden layers using f

+ f

. We use an activation

function to train the complex patterns and take deci-

sions for passing the values for the next layers. The

internal working procedure of this model is explained

in the Algorithm 1.

We provide the list of parameters used for training

with the validation inputs below.

• Data Division: Random (70 training, 30 testing)

• Number of input layer: 18

• Number of hidden layers: 5

• Number of output layers: 4

Algorithm 1: CFBPNN operation.

1: Create a simple connected network with a input

and output unit and Initialize y

2: Add the bias by increasing one by one unit

∑

i=1

∑

3: Initialize R

4: Calculate: f



∑

j=1



∑

i=1



5: while (MSE == MSE

threshold

OR the hidden unit

is more than the given value) do

6: Add the linearly independent units to the net-

work one by one.

7: Select the input unit and calculate the sum of

weight

∑

i=1

from the beginning.

8: Add the bias

∑

for each node, and calculate

the weights by connecting each node.

9: Addition of hidden units till (i − k < −N) and

generate the target units.

10: Calculate MSE

∑

i=1

−

11: end while

12: Calculate the weights

∑

of output layer using

back substitution method.

• Number of neurons in each layer: 10

• Training Function: Trainlm

• Maximum number of epochs: 1000

We use the Levenberg-Marquardt training func-

tion for all the network models. This is the fastest

back-propagation algorithm that updates weight and

bias. Comparatively, this method requires more mem-

ory.

4 RESULTS AND COMPARISON

4.1 Evaluation Metrics

A confusion matrix is the most appropriate tech-

nique to analyze the performance of the classiﬁca-

tion model. This helps to indicate the true and false

classiﬁed cases in the tested model. We have con-

sidered false rate and overall accuracy for evaluating

the model performance given below with mathemati-

cal expressions.

False Negative Rate (FNR): Miss classiﬁed to a

selected class, calculated with the Equation 2.

FNR =

T P + FN

(2)

Accuracy (A): The ratio of correctness for classi-

ﬁed samples to right class represented with the Equa-

tion 3.

Accuracy(A) =

T P + TN

T P + FP + FN + T N

(3)

Towards a Novel Approach for Smart Agriculture Predictability

101

4.2 Results of Prediction Model

Our proposed CFBPNN model is used for mapping

the patterns between input and target values. Various

compositions of threshold functions are used in the

layers with multiple combinations. We have trained

and tested the model for the controller 2 data and pro-

jected the results in Figure 3.

To ﬁnd the efﬁciency of the model and identify

the water level and bacteria effect, we have analyzed

the performance of the proposed model using a confu-

sion matrix. This classiﬁes and denotes the accuracy

ratio and the false rate for each cluster. A multi-class

testing is used to identify the relation of each variable

with the cluster; This analysis shows that our model

provides a 98.9% high accuracy rate and very mini-

mal false rates with 1.2% for each test cluster given in

Figure 3.

Figure 3: Confusion matrix of CFBPNN prediction model.

4.3 Variables Effecting Bacteria Levels

To identify the effect of bacteria, on the plant we have

considered four input variables of cluster 1 which

have a high impact on the existence of bacteria con-

tent. A co-relation plot is used to display the rela-

tionship between all the variables of cluster 1 and the

bacteria as the target variable given in Figure 4.

According to Figure 4, it is to be observed that

variable tuC(electricity level) solar Radiation and

Temperature have a high impact on the creation and

spreading of the bacteria. A notable point solar ra-

diation with a 0.68 ratio, has been a major cause of

bacteria development in the plant.

4.4 Variables Effecting Water Levels

Analysing the water requirement provide limited wa-

ter, to preserve water wastage is the huge task. Con-

sidering this we have examined a cluster which inﬂu-

ence more to predict the water requirement is given in

Figure 5. Variable WVC give the count of Volumetric

Water Content which helps in testing the soil moister,

requirement of water level , and the atmosphere con-

dition. This has a high impact with electric control

variable with 0.88 ratio.

4.5 Comparative Analysis

Focusing the crucial requirement of IoT agriculture,

as to predict the water requirement ratio, best irriga-

tion period, suitable temperature, soil moisture lev-

els, and best suitable crop and many more. Some of

the latest proposals using machine and deep learning

models, for various levels of prediction are discussed

in Table 3. Crop prediction by (Varman et al., 2017)

gained minimum loss with 2.135 with 57.65s a mini-

mum training time compared to all the proposed mod-

els. Soil moisture prediction for providing best nutri-

tion for the plant by (Goap et al., 2018), using SVR

and K-means technique resulted with 96% R-square

value. Weather prediction with MLR gained 99.05 ac-

curacy in identifying the apt temperature need to grow

the plant (Parashar, 2019). The combination of mul-

tiple parameters as soil-moisture, air-humidity, air-

temperature using deep-learning model resulted with

least error rates (Dahane et al., 2020). Other predic-

tion model trained using LSTM resulted with RMSE

3.0 for temperature prediction and 14.55 for humidity

prediction to select a plant for irrigation (Jin et al.,

2020).

Capturing live image and detect the disease, us-

ing SVR, and also predict the weather to trigger ac-

tion for pest control by (Sasi Supritha Devi et al.,

2020). The study focused on image processing with

K-Means clustering approach and resulting with an-

droid application for monitoring the irrigation pro-

cess. Classiﬁcation and quantitative predictions for

various parameters as soil type, crop type and amount

of irrigation required using SVM (Support Vector Ma-

chine), SVR (Support Vector Regression) and Ran-

dom forest 81.6% accuracy (Vij et al., 2020). Predic-

tion of Volumetric soil moisture by (Kashyap et al.,

2021) gained 0.012 RMSE for Sutlej Basin plant-

Ludhiana(Punjab). The model proves to be more ac-

curate as the RMSE value is close to zero. They also

explored various method of prediction amount of wa-

ter saved and irrigation period by controlling the func-

tionality of the irrigation scheduler. Crop prediction

ICSOFT 2023 - 18th International Conference on Software Technologies

102

Figure 4: Variables effecting bacteria levels.

Figure 5: Variables effecting water levels.

based on temperature and rainfall using multi-layer

perceptron a rules-based classiﬁer and JRip decision

table classiﬁer by (Bakthavatchalam et al., 2022). The

model resulted with 98.2% accuracy in prediction

with 8.05s training time.

Out of all the above state of models, proposed

model gave highest accuracy comparatively. The

model has a unique technique for clustering and clas-

sifying the resulted clusters, to predict the feature set.

All the given models are only used to predict a spe-

ciﬁc category, our model divide set of features to cre-

ate a universal clusters and predict parameters based

on requirements.

5 CONCLUSION

Research committee all over is striving hard to en-

hance agriculture productivity by deploying services

provide by IoT technologies. In this paper, we have

discussed a platform and topology which helps to ac-

cess sensor-based data and facilitates farmers to take

the right decision to enhance crop productivity. In

this study we have proposed unsupervised clustering

method SOM to form similar variable groups.

In addition, this paper provides an novel tech-

nique CFBPNN to predict the water requirement

and resulted with 98.8% accuracy in classiﬁcation.

An overview on current technologies, models and

frameworks contributed towards smart irrigation is

Towards a Novel Approach for Smart Agriculture Predictability

103

Table 3: Comparative analysis.

Reference Objective Model Results

(Varman et al., 2017) Predict best-suited crop Long Short Term Memory

(LSTM)

Validation Loss=

2.1354

(Goap et al., 2018) Predict soil moisture Support Vector Regression

(SVR) ,K-means

R-squared=96%

(Parashar, 2019) Predict the weather Multiple Linear Regression

(MLR)

Accuracy=99.05%

(Dahane et al., 2020) Multi parameter prediction LSTM, GRU-based model MSE=0.02

(Jin et al., 2020) Prediction of temperature

and humidity

Gated Recurrent Unit

(GRU), LSTM

RMSE=3.00

(Vij et al., 2020) Multi parameter prediction SVM (Support Vector Ma-

chine), SVR (Support Vec-

tor Regression)and Random

forest

Accuracy=81.6%

(Kashyap et al.,

2021)

Prediction of soil moisture

content

long short-term memory

network

RMSE= 0.012

(Rezk et al., 2021) Crop productivity and

drought predictions

wrapper feature selection,

and PART classiﬁcation

technique

Accuracy: 98.15%.

(Bakthavatchalam

et al., 2022)

Crop prediction DL techniques Accuracy= 98.2%

Proposed model Clustering, classiﬁcation

and prediction

SOM and CFBPNN Accuracy= 98.8%

provided. This research explore various challenges

and and requirements for the better understanding of

smart farming. Furthermore, features playing im-

portant role in predictions are visualized using co-

relation map. Finally, this study provide challenges

faced by agriculture industry and prose smart method

to handle it.

ACKNOWLEDGEMENTS

The work is carried out in the frame of the PREC-

IMED project that is funded under the PRIMA Pro-

gramme. PRIMA is an Art.185 initiative supported

and co-funded under Horizon 2020, the European

Union’s Programme for Research and Innovation.

(project application number: 155331/I4/19.09.18).

REFERENCES

Akhter, R. and Soﬁ, S. A. (2022). Precision agriculture us-

ing iot data analytics and machine learning. Journal of

King Saud University-Computer and Information Sci-

ences, 34(8):5602–5618.

Ayaz, M., Ammad-Uddin, M., Sharif, Z., Mansour, A., and

Aggoune, E.-H. M. (2019). Internet-of-things (iot)-

based smart agriculture: Toward making the ﬁelds

talk. IEEE access, 7:129551–129583.

Bakthavatchalam, K., Karthik, B., Thiruvengadam, V.,

Muthal, S., Jose, D., Kotecha, K., and Varadarajan,

V. (2022). Iot framework for measurement and pre-

cision agriculture: predicting the crop using machine

learning algorithms. Technologies, 10(1):13.

Ben Abdallah, E., Grati, R., and Boukadi, K. (2023). To-

wards an explainable irrigation scheduling approach

by predicting soil moisture and evapotranspiration via

multi-target regression. Journal of Ambient Intelli-

gence and Smart Environments, (Preprint):1–22.

Dahane, A., Benameur, R., Kechar, B., and Benyamina, A.

(2020). An iot based smart farming system using ma-

chine learning. In 2020 International symposium on

networks, computers and communications (ISNCC),

pages 1–6. IEEE.

Geetha Lekshmy, V., Vishnu, P., and Harikrishnan, P.

(2022). Adaptive iot system for precision agriculture.

In Inventive Computation and Information Technolo-

gies, pages 39–49. Springer.

Goap, A., Sharma, D., Shukla, A. K., and Krishna, C. R.

(2018). An iot based smart irrigation management

system using machine learning and open source tech-

nologies. Computers and electronics in agriculture,

155:41–49.

Jimenez, A.-F., Ortiz, B. V., Bondesan, L., Morata, G., and

Damianidis, D. (2021). Long short-term memory neu-

ral network for irrigation management: a case study

from southern alabama, usa. Precision Agriculture,

22(2):475–492.

Jin, X.-B., Yu, X.-H., Wang, X.-Y., Bai, Y.-T., Su, T.-L.,

and Kong, J.-L. (2020). Deep learning predictor for

ICSOFT 2023 - 18th International Conference on Software Technologies

104

sustainable precision agriculture based on internet of

things system. Sustainability, 12(4):1433.

Kashyap, P. K., Kumar, S., Jaiswal, A., Prasad, M., and

Gandomi, A. H. (2021). Towards precision agricul-

ture: Iot-enabled intelligent irrigation systems using

deep learning neural network. IEEE Sensors Journal,

21(16):17479–17491.

Kohonen, T. (1990). The self-organizing map. Proceedings

of the IEEE, 78(9):1464–1480.

Lavanya, G., Rani, C., and GaneshKumar, P. (2020). An au-

tomated low cost iot based fertilizer intimation system

for smart agriculture. Sustainable Computing: Infor-

matics and Systems, 28:100300.

Mehra, M., Saxena, S., Sankaranarayanan, S., Tom, R. J.,

and Veeramanikandan, M. (2018). Iot based hydro-

ponics system using deep neural networks. Computers

and electronics in agriculture, 155:473–486.

Nisha, G. and Megala, J. (2014). Wireless sensor network

based automated irrigation and crop ﬁeld monitoring

system. In 2014 Sixth international conference on ad-

vanced computing (IcoAC), pages 189–194. IEEE.

Ohana-Levi, N., Ben-Gal, A., Peeters, A., Termin, D.,

Linker, R., Baram, S., Raveh, E., and Paz-Kagan, T.

(2021). A comparison between spatial clustering mod-

els for determining n-fertilization management zones

in orchards. Precision Agriculture, 22(1):99–123.

Parashar, A. (2019). Iot based automated weather re-

port generation and prediction using machine learn-

ing. In 2019 2nd International Conference on Intel-

ligent Communication and Computational Techniques

(ICCT), pages 339–344. IEEE.

Pathan, M., Patel, N., Yagnik, H., and Shah, M. (2020).

Artiﬁcial cognition for applications in smart agricul-

ture: A comprehensive review. Artiﬁcial Intelligence

in Agriculture, 4:81–95.

Prabhu, B., Sophia, S., and Mathew, A. (2014). A review of

efﬁcient information delivery and clustering for drip

irrigation management using wsn. A., A Review of

Efﬁcient Information Delivery and Clustering for Drip

Irrigation Management Using WSN.

PRIMA (2023). PRECIMED - Precision Irrigation Manage-

ment to Improve Water and Nutrient Use Efﬁciency

in the Mediterranean Region. https://mel.cgiar.org/

projects/precimed. (Accessed on 02/26/2023).

Reddy, K. S. P., Roopa, Y. M., LN, K. R., and Nandan,

N. S. (2020). Iot based smart agriculture using ma-

chine learning. In 2020 Second International Con-

ference on Inventive Research in Computing Applica-

tions (ICIRCA), pages 130–134. IEEE.

Rezk, N. G., Hemdan, E. E.-D., Attia, A.-F., El-Sayed, A.,

and El-Rashidy, M. A. (2021). An efﬁcient iot based

smart farming system using machine learning algo-

rithms. Multimedia Tools and Applications, 80:773–

797.

Sarker, I. H. (2021). Deep learning: a comprehensive

overview on techniques, taxonomy, applications and

research directions. SN Computer Science, 2(6):420.

Sasi Supritha Devi, Y., Kesava Durga Prasad, T., Saladi, K.,

and Nandan, D. (2020). Analysis of precision agri-

culture technique by using machine learning and iot.

In Soft Computing: Theories and Applications: Pro-

ceedings of SoCTA 2019, pages 859–867. Springer.

Swamynathan, M. (2019). Mastering machine learning

with python in six steps: A practical implementa-

tion guide to predictive data analytics using python.

Apress.

Varghese, R. and Sharma, S. (2018). Affordable smart farm-

ing using iot and machine learning. In 2018 Sec-

ond International Conference on Intelligent Comput-

ing and Control Systems (ICICCS), pages 645–650.

IEEE.

Varman, S. A. M., Baskaran, A. R., Aravindh, S., and

Prabhu, E. (2017). Deep learning and iot for smart

agriculture using wsn. In 2017 ieee international con-

ference on computational intelligence and computing

research (ICCIC), pages 1–6. IEEE.

Vij, A., Vijendra, S., Jain, A., Bajaj, S., Bassi, A., and

Sharma, A. (2020). Iot and machine learning ap-

proaches for automation of farm irrigation system.

Procedia Computer Science, 167:1250–1257.

Towards a Novel Approach for Smart Agriculture Predictability

105