DOI: https://doi.org/10.52756/ijerr.2024.v46.014 Int. J. Exp. Res. Rev., Vol.

46: 177-190 (2024)

Automated Machine Learning Classification Framework to Predict Crop Yield and Detect Pest
Patterns

Gopi R1*, Tamil Selvi M2, Saranraj G3, Nagaraj P4, Parthiban K5 and Ranjith Kumar A6
1
Faculty of Computer Science & Engineering, Dhanalakshmi Srinivasan Engineering College, Perambalur–621212,
Tamil Nadu, India; 2Faculty of Computer Science and Engineering, Roever Engineering College, Perambalur-
621212, Tamil Nadu, India; 3Faculty of Artificial Intelligence and Data Science, Dhanalakshmi Srinivasan College of
Engineering, Coimbatore 105, Tamil Nadu, India; 4Faculty of Artificial Intelligence and Machine Learning, K.
Ramakrishnan College of Engineering, Samayapuram - 621112, Tamil Nadu, India; 5Faculty of Computer Science
and Engineering, Dhanalakshmi Srinivasan College of Engineering, Coimbatore - 641105, Tamil Nadu, India;
6
Faculty of Computer Science and Engineering, Lovely Professional University, Phagwara, Punjab, India
E-mail/Orcid Id:
GR, [email protected], https://orcid.org/0000-0003-4957-1843; TSM, [email protected], https://orcid.org/0009-0003-4642-2807;
SG, [email protected], https://orcid.org/0009-0000-6412-115X; NP, [email protected], https://orcid.org/0009-0007-9438-1973;
PK, [email protected], https://orcid.org/0009-0007-2385-5529; RKA, [email protected], https://orcid.org/0000-0003-4383-9212

Article History: Abstract: Plant disease identification is crucial to food security and agricultural product
Received: 01st
Jun., 2024 availability. Traditional disease diagnosis can be tedious, annoying, and inaccurate. The
Accepted: 18th Dec., 2024 investigation examines how modern machine learning algorithms might improve plant
Published: 30th Dec., 2024 disease diagnostics for efficacy and precision. Despite this, machine learning faces
Keywords: many obstacles, including model training, processing costs, and rising demand for large
Machine Learning, Plant data sets. This study proposes a novel method called Automated Machine Learning
Disease, Classification, Classification Framework (AMLCF) to predict crop yield and detect pest patterns. This
Crop yield, Reliability framework simplifies model selection, hyperparameter adjustment, and feature
engineering for non-experts. The amount of time and computational resources needed
have additionally been greatly reduced. The suggested AMLCF is evaluated on different
How to cite this Article: unique agricultural datasets to validate its plant disease detection versatility. Our
Gopi R, Tamil Selvi M, Saranraj G,
Nagaraj P, Parthiban K and Ranjith Kumar extensive simulation analysis found that AMLCF exceeds existing machine learning
A (2024). Automated Machine Learning
Classification Framework to Predict Crop methods in speed, accuracy, and usability. AMLCF's detailed demonstration shows this;
Yield and Detect Pest Patterns.
International Journal of Experimental
besides predicting plant illnesses, this system can predict crop yield and detect pests.
Research and Review, 46, 177-190. Those findings suggest AMLCF could transform farming. Better plant health
DOI:
https://doi.org/10.52756/ijerr.2024.v46.014 monitoring, early disease identification, and farmer selection could be achieved. The
experimental results show that the proposed AMLCF model increases the accuracy ratio
by 92.6%, computational efficiency analysis by 97.4%, versatility analysis by 98.3%,
user accessibility ratio by 99.1%, and crop health tracking analysis by 94.8% compared
to other existing models.

Introduction it difficult for models to generalize across contexts

Machine learning has revolutionized plant disease (Ahmed and Ganapathy, 2021; Chaudhary et al., 2023).
detection and these algorithms improve diagnostic Plant species, infection stage, and weather can affect
efficiency and accuracy (Gong, 2021; Kaur, 2023; Rao et symptom severity and these models struggle to prove
al., 2024; Prasad and Agniraj, 2024). The first issue is generalisability (Huang et al., 2021). Training requires
that plant diseases are diverse and unpredictable, making big, diverse datasets, which can be difficult to get due to
them hard to control. Plant symptoms are diverse, making unpredictability (Wang, 2023). Many plant diseases share

*Corresponding Author: [email protected] 177

 Int. J. Exp. Res. Rev., Vol. 46: 177-190 (2024)
symptoms (Yang, 2020). This increases false positives through the integration of AI. This study aims to analyze
and negatives and makes discrimination harder, along AI-driven systems for online art classes, online basic
with data quality, soabelling is crucial (Ahmad et al., education assessments, online preschool AI curricula, and
2022). Machine learning models need high-quality, well- online English classes to improve learning results and
annotated datasets (Pei and Wang, 2022). student engagement.
Real-world agricultural applications of machine Sun et al., 2021 developed an AI-enhanced online
learning systems must understand the technology's limits English teaching system that employs deep learning and
(Yang et al., 2023). Portable or field-based computers decision tree algorithms to deliver tailored lessons to each
have storage and processing limitations (Rong, 2021). student. According to the test findings, the method
Because farmers sometimes lack technical know-how, significantly enhances students' learning capacity and the
these devices must be straightforward (Shuliar et al., content's practicality.
2023). Models may be hard to adapt to unique An article by He and Sun (2021) presents AI-CATM,
geographies and environmental variables, affecting an artificial intelligence-driven computer-assisted
performance and illness development (Barakina et al., teaching model designed to collaborate with teachers in
2021). developing personalized art lessons tailored to the unique
Quality and quantity remain issues; gathering large needs of each student. This method substantially
datasets with suitable tags is expensive and complex
enhances study depth, student involvement, and creative
(Fitria, 2021). Poorly trained or chaotic models may not
design compared to traditional methods.
generalize or be accurate. Model training is complicated
The study conducted by Li and Su (2020) employed
and constrained by plant disease variability by species
EM-EW and grey clustering analysis to assess AI-
and habitat. Plant diseases vary significantly, making
integrated online basic education training. The model
generalizations problematic (Yanru, 2021). Due to
memory and processing power requirements, deep improves online teaching and provides insights for AI-
learning may be too expensive for field applications. driven basic education with its practical methods.
Integrating the idea into real agricultural systems needs The article by Sun (2021) integrates "5G +AI" and
many steps, including making it farmer-friendly and holographic technologies into online spoken education
responding to field circumstances. These difficulties must for interactive learning using gesture detection. The
be resolved quickly to improve machine learning-based strategy improves students' comprehension, critical
plant disease diagnosis. thinking, and involvement over traditional methods.
The main contributions of the study are- The research by Su and Zhong (2022) uses problem-
#Designing the AMLCF to employ contemporary based learning (P-L) and social robots to create an
machine learning to quickly and reliably diagnose plant artificial intelligence (AI) curriculum for preschools that
diseases. is based on goals, content, methodologies, and
#The proposed system performs model selection, assessment. By focussing on problem-based learning in
tweaking, and feature engineering to solve machine particular, the method enhances AI literacy and the
learning problems. Complex models and high processing efficacy of learning in young children.
costs are hurdles. Nabarun Dawn et al. (2023) suggested that Artificial
#The experimental results have been performed, and Intelligence, Machine Learning and the Internet of Things
the suggested AMLCF model increases the accuracy (IoT) revolutionize Agriculture. New tools, difficulties,
ratio, computational efficiency analysis, versatility
and the potential future of artificial intelligence (AI) in
analysis, user accessibility ratio and crop health tracking
agriculture are explored in depth in this study. In recent
analysis compared to other existing models.
years, expert systems (ES), the IoT, and Artificial Neural
#This section of the written research paper will
Network (ANN) models may unleash AI's full potential
conclude with the following outline: Section 2 focuses on
in the agricultural sector.
modern machine-learning methods for precise plant
disease identification. Section 3 presents the AMLCF. Hemanta Gogoi et al. (2023) proposed the avian
Section 4 offers a comprehensive appraisal, covering all diversity in the paddy field ecosystem surrounding the
aspects, such as effects and drawing connections to Assam university campus in Silchar during the rainy
previous endeavours. Section 5 displays the results. season. There is a significant correlation between the
amount of wasted rice and rice yields, and birds often eat
Related works newly planted rice seeds. Insect pests abound in paddy
This examination delves into many innovative ways fields, which are a semi-aquatic habitat. Waterfowl,
that can be used to improve educational outcomes wading birds, shorebirds, and other species of waterbirds
DOI: https://doi.org/10.52756/ijerr.2024.v46.014
178
 Int. J. Exp. Res. Rev., Vol. 46: 177-190 (2024)
use these areas as staging grounds for migration and example, the author demonstrates how recent
feeding grounds. We present 95 bird species, representing breakthroughs in image recognition technology can
37 families and 14 orders, from this pilot research carried automate the process of identifying specific individuals in
out in the paddy field environment around the Silchar static images, opening the door to the possibility of
campus of Assam University. systematic monitoring of insect pests.
In (Ilham Amani Rozaini et al., 2023) recommended Sangyeon Lee and Choa Mun Yun (2023) introduced
the bilateral teleoperation with a shared design of master sequential environmental data based on predicting risks
and slave devices for robotic excavators in agricultural of crop pests and diseases using deep learning models.
applications. The operational complexity is greatly The author demonstrated the model's ability to forecast
affected by the dynamic used in this teleoperation system, the risk score of agricultural diseases and pests using
which is a master device that controls the activities of an large-scale public data on strawberry, pepper, grape,
agent device. Therefore, this work aims to improve the tomato, and paprika crops. Its average AUROC was
master-slave algorithm for teleoperation applications that 0.917, indicating strong predictive ability with these
depend on regulating the motions of robot arms. predictions, one may aid in preventing pests or post-
Although the master and slave devices are different in processing. Many different facilities and crops may
size, they are structurally similar. Ensing the benefit from this learning framework and model for
understandable kinematic model that connects these parts predicting crop diseases based on environmental data.
is important for making the robot easy to manage and Sheela (2023) suggested Crop Yield Improvement
user-friendly. The robot arm's forward kinematics, with Weeds, Pest and Disease Detection. The data
computed using Denavit-Hartenberg parameters and augmentation procedure is performed since Deep
transformation matrices, must be used to calculate the Learning performs better with bigger data sets. Various
end effector movement and location. DCNN architectures were used to construct the neural
Abderraouf Amrani et al. (2024) discussed the multi- model, and their accuracy and performance were used to
task learning model for agricultural pest detection from interpret the models using hyperparameter-searching,
crop-plant imagery. This approach uses a joint loss InceptionV3, DenseNet201, Mobilenet, VGG16, and
function, which merges a classification loss with a hyperparameter-tuning on data from agricultural picture
tailored size loss. An image containing an aphid may be sources. When comparing the end value, the tweaked
located using the classification component, and its size InceptionV3 model performed 87.85% better. In contrast,
can be approximated using the customized size loss Mobilenet and VGG16 got 91.85% and 78.71% accuracy,
function. The latter is personalized-made for a more respectively. The DenseNet model outperformed the
precise size estimate to smooth out differences between Hyperparameter Search with a 99.62% accuracy rate.
predicted and measured ground truth sizes. This model is In Md. Akkas Ali et al. (2023) proposed sound
built with a ResNet18 backbone, which makes it very analytics in large agricultural fields for pest detection
flexible and resilient. systems using high-performance-oriented AI-enabled
Kariyanna and Sowjanya (2024) deliberated on IoT. To denoise the pest sound, remove spectral leakage,
theunravellingf use of artificial intelligence in managing convert overlapping to non-overlapping frames, convert
insect pests. Using sophisticated algorithms, AI provides the time to the frequency domain, determine the
a game-changing method for analyzing complex data frequency spectrum, detect the sinusoidal frequency and
patterns gathered from many sources, such as sensors and internal component, and extract the MFCCs feature, the
pictures. Reducing the need for random pesticide proposed method utilized audio pre-processing
applications and making the most of interventions allows techniques in sound analytics, such as HPF, Hann
for precise identification of pests, early diagnosis of window, hop window, FFT, DFT, STFT and the MFCC
problems, and predictive modelling, all of which improve algorithm.
pest management decision-making. In Rashmi Priya Sharma et al. (2023) recommended
Philipp Batz et al. (2023) presented the potential of the Internet of Farm Things-based prediction for crop
image recognition and artificial intelligence for aphid pest infestation using an optimized fuzzy inference
pest monitoring. Using aphids as an example, the system (IoFT-FIS). Using these meteorological
author shows how future technological advances in characteristics, the knowledge base of the proposed fuzzy
systematic monitoring of insect pests, automated inference system is constructed. The multi-objective
individual identification, and intelligent forecasting evolutionary algorithm uses fuzzy criteria to determine an
models might benefit from case data. Using aphids as an appropriate cropping window and breeding circumstances
DOI: https://doi.org/10.52756/ijerr.2024.v46.014
179
 Int. J. Exp. Res. Rev., Vol. 46: 177-190 (2024)
with low pest populations. Using an IEEE 802.15.4 health management. The suggested technique aims at a
wireless IoT sensor network monitoring infrastructure in rapid, accurate diagnosis of plant diseases that strive for
medium grass vegetation, this proposal finds crop-sowing power with ease.
windows based on fuzzy logic with optimum crop The suggested technique presents the AMLCF to
production and lowest insect development. Crops of forecast agricultural yields and identify pest trends
sugarcane and rice are the subjects of current accurately. This framework incorporates state-of-the-art
experiments. This study's experimental setting was a farm machine learning algorithms. It aims to improve
near Gwalior, Madhya Pradesh, India. The field-deployed adaptation across agricultural datasets by focusing on
wireless sensor network gathered soil moisture, rainfall, automated feature selection and model improvement.
temperature, and more data. Computationally efficient, scalable, and easily accessible,
Galiya Anarbekova et al. (2024) discussed the fine- the framework is designed to make it easy for farmers
tuning of artificial neural networks to predict pest and other stakeholders in the agricultural sector to
numbers in grain crops. The transformer reliably shows deploy. The AMLCF offers a centralized platform for
better prediction accuracy in terms of mean squared error. tracking pest activity and crop health, promoting
The effect of several training hyperparameters on proactive management and well-informed decisions in
predicted accuracy, including batch size and epochs, is precision agriculture. To promote sustainable agricultural

Figure 1. Block diagram for Data Collection.

also shown in this paper. Season 2's unusual reactions methods, these contributions tackle important issues in
highlight the impact of certain characteristics on model contemporary agriculture, such as optimizing resources
performance, which is an intriguing development. This and controlling pests promptly.
study adds to the growing knowledge of optimizing Contribution 1: Design of AMLCF
ANNs for precise insect prediction in grain crops, which The following are three stages the recommended
helps create more effective and efficient pest AMLCF goes through to improve detection efficiency
management methods. It is also well-suited for real-world greatly: Auto-preprocessing, segmentation, and feature
applications due to the transformer model's persistent extraction. This simplified process reduces computation
dominance. resources and time needed for plant disease detection.
Moreover, state-of-the-art machine learning approaches
Materials and Methods used in this framework boost classification precision,
The suggested approach uses an AMLCF, or resulting in reliable and accurate identification of various
Automated Machine Learning Classification Framework, plant diseases. Improved crop health management
to improve plant disease detection. AMLCF's use of follows, as well as timely intervention actions.
advanced machine-learning techniques makes it easier for Figure 1 shows the data-gathering procedure used to
the system to capture images and the subsequent identify plant diseases within an AMLCF framework.
categorizing of diseases. The major goal is to increase The initial step involves capturing raw pictures of plants
agricultural production through improved efficiency and using cameras or sensors, which is called image
accuracy of plant disease detection, which helps crop acquisition. Each picture is then annotated to show
DOI: https://doi.org/10.52756/ijerr.2024.v46.014
180
 Int. J. Exp. Res. Rev., Vol. 46: 177-190 (2024)
relevant information, such as the kind of ailing plant, are denoted by 𝑟 −𝑟(𝑎) and 𝑜4 .
after which it is stored in an annotated dataset. 𝑣3 𝑟(𝑎+𝑑)+𝑟(𝑎) 𝑣 𝑣1 2𝑟(𝑎+𝑑)
𝑣
𝑔 + 𝑣4 ℎ𝑟(𝑎+𝑑)−𝑟(𝑎) − 𝑣2
𝑓
2 2
Then, these photos are pre-processed during the image
(2)
processing phase to be better enhanced to find patterns. 𝑣3
For instance, this stage can utilize noise reduction, The equation 2 could be understood as 𝑣2
representing
𝑣4
contrast adjustments or segmentation methods to separate various plant health indicators , and 𝑔 and ℎ
𝑣2
areas of interest depicted in the photos from each other. 𝑣1
representing features extracted 𝑎 + 𝑑 from agricultural
Next comes feature extraction, where certain attributes or 𝑣2
features are recognized and extracted from the processed datasets. Possible representations 𝑟(𝑎) of the model's
pictures. These attributes could be texture, colour and sensitivity to changes in data or circumstances may be
form, among others. These characteristic features are found in the exponents of (𝑎 + 𝑑) and −𝑟(𝑎).
necessary for differentiating between different illnesses in This study's simulation environment was built using
plants. Lastly is classification, where machine learning MATLAB Simulink, a powerful tool for modelling,
methods are employed to sort the images into different simulating, and evaluating dynamic systems. The
illness categories based on the retrieved attributes. The deciding factors were the environment's adaptability in
model trained on the identified dataset to make illness incorporating machine learning models for prediction
predictions in fresh, unlabelled photos. Improved plant tasks and its capacity to manage massive agricultural
health surveillance and early disease diagnosis are made information. Using past data on crop yields, patterns of
possible by this simplified technique, which is made pest infestation, and weather variables like temperature,
possible by AMLCF, which increases the effectiveness humidity, and rainfall made the simulation seem like real-
and reliability of plant disease identification. life agricultural circumstances. Because of the system's
𝑠1 𝑜 𝑟(𝑎+𝑑) + 𝑜2 𝑟 −𝑟(𝑎+𝑑) = 𝑜3 𝑟 𝑟(𝑎) + 𝑜4 𝑟 −𝑟(𝑎) adaptability, seasonality and other regional considerations
(1) might be dynamically important. This study verified the
The Equation 1 may be connected to the suggested suggested framework's accuracy and resilience in this
Automated Machine Learning Classification Framework controlled setting, which minimized the time and money
𝑠1 as follows factors (𝑎 + 𝑑) or elements that are needed for actual field trials.
tweaked 𝑜2 throughout −𝑟, the procedure of machine Contribution 2: Implementation of Machine Learning
learning is given by 𝑎 + 𝑑 and 𝑟(𝑎), but the scores or Model selection, hyperparameter tweaking, and
indices modified to maximize the reliability of the model feature engineering are just a few of the frequent

Figure 2. Process of Automated Machine Learning for Plant Disease

deduction.

DOI: https://doi.org/10.52756/ijerr.2024.v46.014
181
 Int. J. Exp. Res. Rev., Vol. 46: 177-190 (2024)
machine-learning difficulties the AMLCF automates. The Equation 4 agreement with the suggested Automated
barrier to entry is lowered, and complex machine learning Machine Learning Classification Framework 𝑁(𝐹) .
methods are accessible to non-experts because of this Accuracy of disease identification analysis, such as
simplicity. More people will use and put the framework 1
𝑔
√ℎ − 𝑟𝑙, and 𝜕(𝑟 − 𝑗𝑓) reflect various model and data
to use in the field because of how easy it is to use, even
attributes that impact 𝜏 the learning process, whereas
for those without much technical training, which is great 𝑟
−ℎ𝑠+
for the agricultural industry. 𝑖 𝑓𝑟 denotes the model's performance metrics.
The whole procedure of using machine learning Simplifies difficult processes ∋ for enhanced plant
𝑟
technology to identify plant diseases is shown in Figure disease detection by balancing computing efficiency
𝑓𝑟
2. The first step is to get photos from field crops,
and accuracy.
specifically from leaves, using either cameras or sensors.
Contribution 3: Demonstrate Real-World Benefits
You may find these pictures in the Leaf Image Database.
and Versatility
Image pre-processing improves the quality of the photos
The AMLCF's flexibility in effectively identifying
by using cropping filters and separation algorithms,
many plant diseases is shown by testing it on diverse
which then isolate the leaf region. This phase is critical
agricultural datasets, showcasing its real-world
for precise disease identification because it zeroes in on
application. The framework has other uses beyond only
the leaf and eliminates extraneous background data. Data
diagnosing diseases; it may identify pests and monitor the
Splitting is the next step after pre-processing the photos;
condition of crops. The impressive results in many
it involves dividing the dataset into three parts: training,
settings highlight its ability to transform farming
validation, and testing. This guarantees accurate model
methods, leading to better yields and more efficient
validation and training as well as reliable performance
operations. The AMLCF's versatility makes it a useful
evaluation.
tool in many agricultural settings.
To train a machine learning model, one uses training
In Figure 3, machine learning for plant disease
and validation datasets. Here, the model is trained to
categorization is displayed. The Dataset has to be
recognize characteristics linked to different plant diseases
expanded as the first step so that the model can
in the training set, and its accuracy is adjusted with the
accommodate more data. To enhance the images further,
help of the validation set. The trained model's
Image Filtering is done on the dataset, removing noise
performance is evaluated using the Test Set. In this stage,
and other irrelevant information. Attributes Selection is
the model's accuracy in disease classification is tested
the next stage in illness classification that comes after
with fresh, previously unseen photos of plants. Lastly,
image filtering to determine which pixels in an image are
Plant Leaf Classification employs the performance-
important. This processing stage determines what leaf
assessed model, which enables precise disease
traits indicate what diseases; hence, it’s crucial for the
identification for efficient plant surveillance and early
model's reliability. After this, datasets are separated into
disease treatment. To improve the accuracy and
two parts, namely, training and test sets. Training data
efficiency of plant disease diagnoses, this AutoML
trains a machine learning model used by models, while
approach simplifies the workflow.
test data assesses its performance in terms of accuracy
𝜗1 𝜗2 𝜗1 𝜗2
𝜎3 ( + ) = 𝜎1 ( − )=0 and generalizability. This ensures that our model can
√1+𝛽 √1−𝛽 √1+𝛽 √1−𝛽
generalize well even on unknown samples.
(3) The above approach gives us a Machine Learning
This is equation 3, seen as a balanced model inside algorithm that trains on our training data and finds
(AMLCF), where various parameters (θ₁, θ₂) are features associated with various diseases of plants. The
standardized and changed according to a factor (β) to accuracy and efficiency of such a trained model are then
attempt to reach an ideal state. Machine learning relies on determined when applied to test data. A trained model is
this equilibrium to achieve high accuracy and efficiency; applied to predict three diseases: brown spots, leaf
it improves performance in agricultural applications like bacteria blight, and leaf moulds. Like any other machine-
plant disease detection by ensuring features are learning technique, this allows future detection of three
appropriately scaled and weighted. crop sicknesses: brown spots, leaf bacteria blight, and
𝑟
1 ∋ −ℎ𝑠+𝑓𝑟 leaf Molds.
𝑁(𝐹) = √ℎ − 𝑟𝑙/𝜕(𝑟 − 𝑗𝑓) = 𝜏 ∫0 𝑖
𝑔 ∑𝑠𝑘=1 𝛼𝑙 + 𝑑𝑙+𝑟 − 𝑠𝑦𝑡+𝑘 > 0, ∝𝑞 + 𝑀𝑅 (5)
(4) A representation of the classification model's

DOI: https://doi.org/10.52756/ijerr.2024.v46.014
182
 Int. J. Exp. Res. Rev., Vol. 46: 177-190 (2024)
threshold-based decision-making 𝑘 = 1 process when variables 𝑎. 𝑒. 𝑥 and 𝜎 𝑖, guarantee the implementation of
Equation 5 is satisfied. The parameters or weights that customized solutions 𝑁 for different agricultural datasets.
have been learnt, 𝛼𝑙 and 𝑑𝑙+𝑟 , aspects or data elements The AMLCF, a computerized machine learning
from the dataset, 0, ∝𝑞 and 𝑠𝑦𝑡+𝑘 , computational classification framework, is shown in Figure 4. The data
efficiency analysis and a margin or regularization term, are from the Agriculture Crop Images Kaggle Dataset.
𝑀𝑅, are all presented this time. The Crop images dataset includes 40 or more images of
𝐺𝑖𝑅 (𝑢𝑅 (𝑥)) = 𝐺𝑖 (𝑢𝑅 (𝑥)) + 𝑎. 𝑒. 𝑥 ∈ 𝜏, 𝜎 𝑖 = every agricultural crop, including sugarcane, rice, jute,
1, … , 𝑁 (6) wheat, and maize. There are more than 160 enhanced

Figure 3. Plant disease classification using machine learning.

Figure 4. The overall structure of AMLCF.

The Equation 6 may be linked to the offered 𝐺𝑖𝑅 . To crop images in each class in the dataset (kag2). Among
improve the algorithm for improved performance 𝑢𝑅 (𝑥), the augmentation features are the ability to rotate, flip
this change is like application versatility analysis horizontally, and shift vertically. Images of each class
hyperparameter tweaking 𝐺𝑖 and feature engineering. The show them at various stages of their life cycle, from
model's thresholds or limitations are represented by the above to below and from aerial to ground level.
DOI: https://doi.org/10.52756/ijerr.2024.v46.014
183
 Int. J. Exp. Res. Rev., Vol. 46: 177-190 (2024)
Improving the accuracy of crop image classification for 5 using fresh data (𝑡) demonstrates the system's capability
agricultural crop types: wheat, rice, sugarcane, maize, for fluid adaptation 𝑖 = 1, … , 𝑁 and refining.
and jute. According to its simplified and automated
First, cameras or sensors are used to take pictures of methodology, the suggested AMLCF technique shows
the plant's leaves, known as the Input Image Leaf. After substantial gains in plant disease diagnosis. Beginning
that, it goes on to pre-processing, where contrast with image capture, the approach consists of many fine-
adjustment and noise reduction are used to improve the tuned phases for accuracy and efficiency: pre-processing,
photos. After the pre-processing, the segmentation segmentation, extracting features, and classification.
procedure is carried out to separate the leaf area from the Testing the framework on various agricultural datasets
backdrop. This ensures that the emphasis is on the correct has shown that the approach outperforms conventional
area. Next, Feature Extraction finds the leaves' unique methods. In addition to detecting plant illnesses, the
characteristics, such as their form, colour and texture, AMLCF helps monitor crop health and manage diseases
important for differentiating between healthy and sick early on. Intending to keep crops healthy, this method is
leaves. The next step, Classification, uses the extracted useful for farmers and experts in agriculture as it
data to classify the leaves as healthy or sick using guarantees efficient computing, high accuracy, and user
machine learning algorithms. accessibility.
After completing the classification, the system creates
Infected and Healthy Images to validate the findings Result and Discussion
visually. The Analysis of Disease section delves further Accuracy, computational efficiency, application
into the recognized disorders to better understand them adaptability, user accessibility, and crop health tracking
and record our findings. Reliable disease detection is are some metrics used to assess the efficacy of advanced
achieved using Performance Measures to evaluate the machine learning algorithms in plant disease diagnosis.
classification model's accuracy and efficiency. The The MATLAB Simulink environment, a flexible tool for
computational efficiency of the framework is maximized modelling and simulating dynamic systems, was used to
by reducing processing time and resource consumption. perform the simulation tests. The capacity to include
To facilitate successful interaction between the system machine learning models with agricultural datasets and
and farmers and agricultural specialists, User the simulator's reliability in handling time-series data
Accessibility guarantees that the system is were the deciding factors in its selection. A time step of
straightforward to use. Crop health tracking is another 0.01 seconds and a simulation length of 10,000 seconds
area where AMLCF has found use. This technology were set up for the experiment. The environmental
allows for the constant monitoring and control of plant elements included temperature (20-35°C) and humidity
health, which helps with improved farming methods and (40-60%), while the crop-specific variables were growth
the early diagnosis of diseases. rate and insect infestation thresholds. These parameter
𝐼𝑎 = [0, 𝑎] ∩ 𝑇 = {𝑡 ∈ 𝑇 ∶ 0 ≤ 𝑡 ≤ 𝑎} = [0, 𝑎]𝑇 sets were evaluated against available literature to
(7) guarantee realism and derived from experimental
The Equation 6 could help reduce computational findings. The system allowed for a comprehensive
resources and improve model efficiency 𝐼𝑎 and accuracy evaluation of the efficacy of agricultural production
[0, 𝑎] by selecting time-specific features or data points ∩ prediction and pest detection by creating a controlled
𝑇 for model training. Improved disease diagnosis simulation environment that successfully mimicked real-
alongside 𝑡 ∈ 𝑇 User accessibility analysis is possible world environments.
because of 𝑡 ability to handle and evaluate 𝑎 the temporal In Figure 5 above, controlling plant diseases
features [0, 𝑎] of agricultural information 𝑇 by efficiently requires this ability; CNNs and deep learning
concentrating on this subset. are two methods that have substantially improved
𝑓(𝜎(𝑡))−𝑓(𝑠) diagnostic accuracy. This has been achieved by carefully
𝑓 ∆ (𝑡) = lim 𝜎(𝑡)−𝑠
𝑖 = 1, … , 𝑁
𝑠→𝑡
examining patterns in botanical images. Despite their
(8)
∆
ability to efficiently distinguish between different
This is the equation 8 function 𝑓 about a timescale
diseases, the effectiveness of these systems is greatly
(𝑡). By seeing 𝑓(𝜎(𝑡)) as the process (𝑓) of adaptive reliant on the diversity and quality of the training data
feature selection lim and model tuning over time. The utilized. It is critical to obtain high-quality, adequately
𝑠→𝑡
enhancer of the function 𝑓(𝑠) crop health tracking tagged datasets to train an accurate model that produces
analysis as AMLCF changes and modifies its models 92.6%. Furthermore, thorough model tweaking and
DOI: https://doi.org/10.52756/ijerr.2024.v46.014
184
 Int. J. Exp. Res. Rev., Vol. 46: 177-190 (2024)
validation are required to reduce the number of mistakes. learning models and CNNs generally need a lot of
AMLCFs can improve accuracy; these frameworks memory and processing power yet produce very accurate
streamline model selection and hyperparameter tuning, results. In particular, when working with massive
making these procedures more user-friendly and datasets, the complicated computations needed to train
effective. There has been a lot of improvement; however, these models can drive up the associated costs and
there are still problems like data unpredictability and the increase the implementation duration by 97.4%. The
strain on computer resources. Consequently, constant development of AMLCFs is a response to these
improvement and updating are necessary. Nevertheless, it problems. These frameworks can make procedures like
is crucial to address these concerns to keep performance model selection and hyperparameter tuning easier.
consistent across various agricultural situations, even Further, these frameworks expedite model deployment
while strong machine learning algorithms greatly while reducing the amount of computational labour
improve disease identification accuracy. needed. Using AMLCFs will make getting optimal

Figure 5. Accuracy of Disease Identification Analysis.

Figure 6. Computational Efficiency Analysis.

In Figure 6 above, analyzing the computational performance with fewer resources a lot easier.
efficiency of potential methods for plant disease Simplifying and automating these processes will make
diagnostics is a crucial part of evaluating advanced this possible. Hardware efficiency and fine-tuned
machine learning algorithms. Techniques like deep algorithms are prerequisites for capable control of
DOI: https://doi.org/10.52756/ijerr.2024.v46.014
185
 Int. J. Exp. Res. Rev., Vol. 46: 177-190 (2024)
resource utilization. This is because, despite essential to consider the user's accessibility initially.
advancements, deep learning algorithms can still retain Technical proficiency and specialized knowledge are
very high computing demands. Finding an appropriate frequently required to apply and interpret new approaches
balance between computational economy and efficiency that offer high accuracy, such as CNNs. Because they
is crucial for real-world applications in agricultural streamline the otherwise convoluted procedures of model
settings, even when modern approaches guarantee a high selection, hyperparameter tuning, and feature
degree of accuracy. engineering, AMLCF can alleviate this problem. This

Figure 7. Application Versatility Analysis.

In Figure 7 above, the wide range of applications of makes these tactics more accessible to those who may not
modern machine learning algorithms in plant disease have a strong background in technology. By automating
identification demonstrates their versatility. Methods like formerly manual procedures, AMLCFs lessen the
CNNs prove their adaptability by detecting numerous demand for specialized skills and facilitate their
plant diseases in various environments and with different incorporation into real-world agricultural contexts. This
species. This proves that CNNs can discriminate between paves the way for their usage to spread, producing 99.1%.
various plant diseases with high accuracy. The availability of user-friendly support tools and
AMLCFs enhance this adaptability by making model interfaces is crucial to ensuring farmers and agricultural
modification easier and facilitating rapid adaptation to professionals can effectively utilize new technologies.
varied datasets and illness types. In addition to helping Although some success has been had, further work is
with disease identification, these frameworks could be required to make the design and support even more user-
useful for other agricultural tasks, including estimating friendly. Despite some success, this remains the case. By
crop yields and preventing pest infestations. Because of taking this route, we can make sure that innovative
the adaptability of modern machine learning algorithms, machine-learning methods are applied to farming in an
they may be used for a broad range of agricultural tasks, approachable and beneficial way.
allowing for comprehensive plant health monitoring, In Figure 9, advanced machine learning methods
which produces 98.3%. To ensure consistent performance greatly improve the capacity to monitor crop health by
across a wide number of applications and conditions, it is allowing for the rapid and precise diagnosis of diseases.
necessary to continually improve and validate it, even if it Methods like CNNs do an excellent job of analyzing
is adaptable. This is the sole method to guarantee that plant images, which can be used to detect disease signs in
there will be ongoing development. By offering complete plants at an early stage. This enables the proactive
solutions to numerous problems related to plant disease management of crop health. The method is further
control, these strategies may cause an unprecedented shift optimized with AMLCFs, and improved access to
in agricultural operations. The methods' broad adoption advanced disease diagnostics results from these
demonstrates their promise and highlights their ability to frameworks' efforts streamline model deployment and
revolutionize agricultural operations. reduce processing requirements. Farmers can get real-
In Figure 8, when applying modern machine learning time alerts and information about possible problems by
algorithms to the problem of plant disease diagnosis, it is incorporating these techniques into crop health
DOI: https://doi.org/10.52756/ijerr.2024.v46.014
186
 Int. J. Exp. Res. Rev., Vol. 46: 177-190 (2024)
monitoring systems, which produce 94.8%. Because of inevitably be obstacles that must be overcome. When
this, farmers can now control and intervene with their considering all factors, crop health management benefits
crops at the perfect moment. Additionally, this substantially from applying advanced machine learning
preventative measure encourages better crop treatment algorithms. Table 1 shows the performance analysis.
decisions and resource allocation and improves the Despite these advancements, complications like data
accuracy of illness diagnosis. variability and computational resource demands persist.

Figure 8. User Accessibility Analysis.

Figure 9. Crop Health Tracking Analysis.

Table 1. Performance Analysis.

Parameter AMLCF AI-OETS AI-CATM EM-EW P-L
Accuracy Ratio (%) 92.6 85.4 88.2 83.7 84.9
Computational Efficiency (%) 97.4 89.3 91.1 87.6 88
Versatility Analysis (%) 98.3 90.2 92.5 89 91.3
User Accessibility Ratio (%) 99.1 92.4 94 91.2 92
Crop Health Tracking (%) 94.8 87.5 89.8 86.2 88
Ongoing issues, such as data quality and model This highlights the critical need for ongoing improvement
flexibility, must be addressed if crop health tracking to ensure efficient and consistent application in various
remains accurate and helpful across various agricultural agricultural contexts.
contexts. Despite potential advantages, there will
DOI: https://doi.org/10.52756/ijerr.2024.v46.014
187
 Int. J. Exp. Res. Rev., Vol. 46: 177-190 (2024)
Conclusion pest detection system using sound analytics in large
The following section of the paper describes how agricultural field. Microprocessors and
advanced machine-learning algorithms for plant disease Microsystems, 103, 104946.
identification can revolutionize the field. Traditional https://doi.org/10.1016/j.micpro.2023.104946
plant disease identification takes time and effort, reducing Amrani, A., Diepeveen, D., Murray, D., Jones, M. G. K.,
food safety and agricultural output. The research solves & Sohel, F. (2024). Multi-task learning model for
the issues completely. An AMLCF is developed to agricultural pest detection from crop-plant imagery:
simplify feature engineering, model selection, and A Bayesian approach. Computers and Electronics in
hyperparameter tuning. Advanced machine learning Agriculture, 218, 108719.
methods become more accessible and efficient. AMLCF https://doi.org/10.1016/j.compag.2024.108719
exceeds standard accuracy, speed, and usability Anarbekova, G., Ruiz, L. G. B., Akanova, A., Sharipova,
approaches when applied to agricultural datasets. This S., & Ospanova, N. (2024). Fine-Tuning Artificial
diagnostic paradigm helps agriculture anticipate crop Neural Networks to Predict Pest Numbers in Grain
yields, identify pests, and diagnose plant diseases. These Crops: A Case Study in Kazakhstan. Machine
findings suggest that AMLCF could revolutionize Learning and Knowledge Extraction, 6(2), 1154–
agriculture by improving crop health monitoring, early 1169. https://doi.org/10.3390/make6020054
disease detection, and farmer decision-making. The fact Barakina, E. Y., Popova, A. V., Gorokhova, S. S., &
that AMLCF can reduce computer expenses and simplify Voskovskaya, A. S. (2021). Digital Technologies
operations makes it suitable for modern agriculture. This and Artificial Intelligence Technologies in
improves sustainable and effective agriculture, and an Education. European Journal of Contemporary
AMLCF tool addresses crucial plant disease control Education, 10(2), 285-296.
challenges. The experimental results show that the https://doi.org/10.13187/ejced.2021.2.285
proposed AMLCF model increases the accuracy ratio by Batz, P., Will, T., Thiel, S., Ziesche, T. M., & Joachim,
92.6%, computational efficiency analysis by 97.4%, C. (2023). From identification to forecasting: the
versatility analysis by 98.3%, user accessibility ratio by potential of image recognition and artificial
99.1%, and crop health tracking analysis by 94.8% intelligence for aphid pest monitoring. Frontiers in
compared to other existing models. Incorporating satellite Plant Science, 14.
images and data from IoT sensors for real-time https://doi.org/10.3389/fpls.2023.1150748
monitoring, the framework may be enhanced to include Chaudhary, Yashi, & Pathak, H. (2023). MCIP: Mining
more varied information from other geographies and Crop Image Data On pyspark data frame Using
crops in future research. Additionally, using advanced Feature Selection and Cluster Based Techniques. Int.
explainable AI approaches may make the model J. Exp. Res. Rev., 34(Special Vol.), 106-119.
predictions more interpretable, which is great for farmers https://doi.org/10.52756/ijerr.2023.v34spl.011
and agricultural professionals since it allows them to Dawn, N., Ghosh, T., Ghosh, S., Saha, A., Mukherjee, P.,
make better decisions. Sarkar, S., Guha, S., & Sanyal, T. (2023).
Implementation of Artificial Intelligence, Machine
Conflict of Interest
Learning, and Internet of Things (IoT) in
The authors declare that there is no conflict of interest.
revolutionizing Agriculture: A review on recent
References trends and challenges. International Journal of
Ahmad, S. F., Alam, M. M., Rahmat, M. K., Mubarik, M. Experimental Research and Review, 30, 190–218.
S., & Hyder, S. I. (2022). Academic and https://doi.org/10.52756/ijerr.2023.v30.018
administrative role of artificial intelligence in Fitria, T. N. (2021, December). Artificial intelligence
education. Sustainability, 14(3), 1101. (AI) in education: Using AI tools for teaching and
https://doi.org/10.3390/su14031101 learning process. In Prosiding Seminar Nasional &
Ahmed, A. A. A., & Ganapathy, A. (2021). Creation of Call for Paper STIE AAS, 4(1), 134-147.
automated content with embedded artificial https://prosiding.stie-
intelligence: a study on learning management system aas.ac.id/index.php/prosenas/article/view/106
for educational entrepreneurship. Academy of Gogoi, H., Purkayastha, J., & Roychoudhury, S. (2023).
Entrepreneurship Journal, 27(3), 1-10. Avian diversity in the paddy field ecosystem
Ali, Md. A., Dhanaraj, R. K., & Nayyar, A. (2023). A surrounding the Assam University campus in Silchar
high-performance-oriented AI-enabled IoT-based during the rainy season. International Journal of
DOI: https://doi.org/10.52756/ijerr.2024.v46.014
188
 Int. J. Exp. Res. Rev., Vol. 46: 177-190 (2024)
Experimental Research and Review, 34(Special Experimental Research and Review, 42, 100-110.
Vol.), 120–137. https://doi.org/10.52756/ijerr.2024.v42.009
https://doi.org/10.52756/ijerr.2023.v34spl.012 Rong, L. (2021). Design of Ideological and Political
Gong, Y. (2021). Application of virtual reality teaching Multimedia Network Teaching Resources
method and artificial intelligence technology in Integration System Based on Wireless Network.
digital media art creation. Ecological Informatics, Scientific Programming, 2021(1), 4293771.
63, 101304. https://doi.org/10.1155/2021/4293771
https://doi.org/10.1016/j.ecoinf.2021.101304 Rozaini, I. A., Ahmad Zakey, N. E. N., Mohd Zaman, M.
He, C., & Sun, B. (2021). Application of artificial H., Ibrahim, M. F., Mustaza, S. M., & Mohamed
intelligence technology in computer-aided art Moubark, A. (2023). Bilateral Teleoperation with a
teaching. Computer-Aided Design and Applications, Shared Design of Master and Slave Devices for
18(S4), 118-129. Robotic Excavators in Agricultural Applications.
https://doi.org/10.14733/cadaps.2021.S4.118-129 International Journal of Experimental Research and
https://www.kaggle.com/datasets/aman2000jaiswal/agric Review, 35, 119–127.
ulture-crop-images?resource=download https://doi.org/10.52756/ijerr.2023.v35spl.011
Huang, J., Saleh, S., & Liu, Y. (2021). A review on Sharma, R. P., Dharavath, R., & Edla, D. R. (2023).
artificial intelligence in education. Academic Journal IoFT-FIS: Internet of farm things based prediction
of Interdisciplinary Studies, 10(3). for crop pest infestation using optimized fuzzy
https://doi.org/10.36941/ajis-2021-0077 inference system. Internet of Things, 21, 100658.
Kariyanna, B., & Sowjanya, M. (2024). Unravelling the https://doi.org/10.1016/j.iot.2022.100658
use of artificial intelligence in management of insect Shuliar, V., Shkurko, V., Polukhtovych, T., Semeniako,
pests. Smart Agricultural Technology, 8, 100517. Y., Shanaieva-Tsymbal, L., & Koltok, L. (2023).
https://doi.org/10.1016/j.atech.2024.100517
Using Artificial Intelligence in Education. BRAIN.
Kaur, P. (2023). Performance and Accuracy
Broad Research in Artificial Intelligence and
Enhancement During Skin Disease Detection in
Neuroscience, 14(3), 516-529.
Deep Learning. Int. J. Exp. Res. Rev., 35, 96-108.
https://doi.org/10.18662/brain/14.3/488
https://doi.org/10.52756/ijerr.2023.v35spl.009
Su, J., & Zhong, Y. (2022). Artificial Intelligence (AI) in
Lee, S., & Yun, C. M. (2023). A deep learning model for
early childhood education: Curriculum design and
predicting risks of crop pests and diseases from
future directions. Computers and Education:
sequential environmental data. Plant Methods, 19(1).
Artificial Intelligence, 3, 100072.
https://doi.org/10.1186/s13007-023-01122-x
https://doi.org/10.1016/j.caeai.2022.100072
Li, M., & Su, Y. (2020). Evaluation of online teaching
Sun, X. (2021). 5G joint artificial intelligence technology
quality of basic education based on artificial
intelligence. International Journal of Emerging in the innovation and reform of university English
Technologies in Learning (iJET), 15(16), 147-161. education. Wireless Communications and Mobile
https://doi.org/10.3991/ijet.v15i16.15937 Computing, 2021(1), 4892064.
Pei, Z., & Wang, Y. (2021). Analysis of computer-aided https://doi.org/10.1155/2021/4892064
teaching management system for music appreciation Sun, Z., Anbarasan, M., & Praveen Kumar, D. J. C. I.
course based on network resources. Computer-Aided (2021). Design of online intelligent English teaching
Design and Applications, 19(S1), 1-11. platform based on artificial intelligence techniques.
https://doi.org/10.14733/cadaps.2022.S1.1-11 Computational Intelligence, 37(3), 1166-1180.
Prasad, P., & Agniraj, S. (2024). Cross-Domain https://doi.org/10.1111/coin.12351
Adaptation Techniques for Robust Plant Disease Wang, Y. (2020). Analysis on the construction of
Detection: A DANN-CORAL Hybrid ideological and political education system for
Approach. International Journal of Experimental college students based on mobile artificial
Research and Review, 42, 68-84. intelligence terminal. Soft Computing, 24(11), 8365-
https://doi.org/10.52756/ijerr.2024.v42.007 8375. https://doi.org/10.1007/s00500-020-04932-6
Rao, K., Devi, J., Anuradha, Y., G, K., Kumar, M., & Yang, H., Shankar, A., & Velliangiri, S. (2022). Artificial
Rao, M. S. (2024). Enhancing Liver Disease intelligence-enabled interactive system modeling for
Detection and Management with Advanced Machine teaching and learning based on cognitive web
Learning Models. International Journal of
DOI: https://doi.org/10.52756/ijerr.2024.v46.014
189
 Int. J. Exp. Res. Rev., Vol. 46: 177-190 (2024)
services. International Journal of e-Collaboration 146-160. https://doi.org/10.3991/ijet.v15i22.18199
(IJEC), 19(2), 1-18. YanRu, L. (2021). An artificial intelligence and machine
https://doi.org/10.4018/ijec.316655 vision-based evaluation of physical education
Yang, R. (2020). Artificial intelligence-based strategies teaching. Journal of Intelligent & Fuzzy Systems,
for improving the teaching effect of art major 40(2), 3559-3569.
courses in colleges. International Journal of https://doi.org/10.3233/JIFS-189392
Emerging Technologies in Learning (IJET), 15(22),

How to cite this Article:

Gopi R, Tamil Selvi M, Saranraj G, Nagaraj P, Parthiban K and Ranjith Kumar A (2024). Automated Machine Learning Classification
Framework to Predict Crop Yield and Detect Pest Patterns. International Journal of Experimental Research and Review, 46, 177-190.
DOI : https://doi.org/10.52756/ijerr.2024.v46.014

DOI: https://doi.org/10.52756/ijerr.2024.v46.014
190