Predictive Modelling of Agricultural Factors to Maximize Crop Yield

Anvesha Nayak, Pramathi Vummadi, Apoorva Raj, Nasam Saimani and Suresh Jamadagni

Department of Computer Science and Engineering, PES University, 100 Feet Ring Road, Bengaluru, India

Keywords: Rice, Paddy, Crop Yield Maximization, Evapotranspiration Forecasting, Surface Runoff Prediction,

Irrigation Management, Pest and Disease Detection.

Abstract: Crop yield prediction and factor analysis are methods through which technology can be utilized to improve

the quality of current Agricultural practices. This study focuses on improving crop yields based on different

factors and ascertaining how climate change affects these factors and their prediction. The aim is to create a

tool for farmers to practice precision agriculture and to be made aware of what controllable factors can lead

to better yield. The study proposes a three-step methodology for this process. First, we will analyse past

years' data and also take into consideration the impact of climate change to know how this relates to these

variables as well as crop yield. Secondly, we suggest some spatial variable management practices that could

improve the overall agricultural output. Along with that, preventative measures to ensure crop safety are

also suggested. Regular updates on these spatial variables will play an important role in helping the farmer

make key decisions during the life cycle of the crop. Finally, in the third step of this process, we aim to

perform anomaly analysis on pests, weeds, diseases, and climatic anomalies, and suggest relevant

countermeasures to the farmer.

1 INTRODUCTION

Agriculture is a fundamental part of human survival

and also one of the key pillars of the world

economy. However, agricultural economic growth

faces an unprecedented challenge in the form of

climate change. We take a case study of paddy

grown predominantly in Asia, specifically India.

Rice is the staple food for more than half the

population of India and one of the critical crops for

the nation’s economy. However, rice cultivation

depends heavily on climate variables such as

temperature, precipitation, etc. Worsening

conditions such as abrupt rises in temperature and

irregular rainfall make optimizing crop yield harder.

Anomalies such as floods, pests, and diseases also

add a layer of unpredictability. This paper aims to

use various datasets related to crop growth,

agricultural factors, weather etc., to study and

understand the relationships between agricultural

productivity and climate variables in India. We also

segregate the entire crop growth cycle into the

different stages and aim to provide suggestions for

each stage based on the output of the machine

learning model. Image analysis models are used to

identify anomalies and diseases while providing

suggestions for tackling them. This paper aims to

identify effective methods to improve the resilience

and sustainability of farming practices through early

detection and timely prevention.

2 RELATED WORK

Crop yield prediction particularly concerning the

climate impacts on wheat production highly features

ensemble models and random forest regression.

Ensemble models, particularly Random Forest

Regressor (RFR), have been gaining increasing

usage in crop yield prediction since they can manage

large complex data and can pick up the nonlinear

relationship between climate variables and yield.

The ensemble model increases the accuracy and

minimizes bias through the combination of various

machine learning insights, thus becoming robust for

high-dimensional, nonlinear climate data (Satpathi et

al., 2023). RFR is highly effective for regional-scale

predictions and feature selection, which minimizes

the risk of overfitting (Breiman, 2001; Pang et al.,

2022). Combining RFR with other models, such as

artificial neural networks or boosted trees, improves

the reliability of predictions under various climate

1360

Nayak, A., Vummadi, P., Raj, A., Saimani, N. and Jamadagni, S.

Predictive Modelling of Agricultural Factors to Maximize Crop Yield.

DOI: 10.5220/0013369400003890

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 17th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2025) - Volume 3, pages 1360-1367

ISBN: 978-989-758-737-5; ISSN: 2184-433X

scenarios. These ensemble-based approaches are

critical in developing climate-resilient agricultural

solutions and managing climate-yield complexities

(Iqbal et al., 2024).

An important aspect of optimization of

agricultural processes is irrigation management.

Irrigation management in turn is heavily dependent

on weather prediction models. The citation (Teixeira

et al., 2024) proposed a combined approach that

integrates Long Short-Term Memory (LSTM) neural

networks with Genetic Algorithms (GA) to forecast

short- and medium-term weather conditions in the

Douro area.

Another weather component that impacts

irrigation management is Evapotranspiration (ETo).

A number of machine learning (ML) and deep

Learning (DL) methods have been proposed to

forecast ETo for improving agricultural productivity.

One paper (Kadkhodazadeh et al., 2022) proposed a

novel approach to predict ETo under climatic

changes using a combination of historical ETo data

and meteorological data through use of regression

models. The study (Granata & Di Nunno, 2021)

explored use of ensemble models on two types of

recurrent neural networks (RNNs), Long Short-Term

Memory (LSTM) and Nonlinear Autoregressive

Network with Exogenous Inputs (NARX) and did a

comparative study on effectiveness in different

climatic conditions (subtropical and semi-arid). The

findings indicated that the LSTM models

outperformed the NARX models in the subtropical

climate, whereas the inverse was true for the semi-

arid climate. Another study (Yin et al., 2020)

addressed the issue of scarce meteorological data for

ETo forecasting through the creation of a hybrid Bi-

directional Long Short-Term Memory (Bi-LSTM)

model. The study highlights the importance of

considering local climate conditions and data

availability when selecting the appropriate

modelling approach.

Prediction of another component, surface runoff,

helps optimize irrigation scheduling and in reducing

water waste due to overirrigation. One paper (Gauch

et al., 2021) presented a novel MTS-LSTM model

for forecasting rainfall-runoff. This method was

developed to predict extreme flooding events at

various time scales, addressing the challenge of

accurately forecasting daily and short-term incidents

that occur more frequently than regular daily

forecasts.

For anomaly detection, recent studies proposed

using CNNs to label rice pests and diseases

strategically by using various advanced architectures

and methods to achieve high accuracy in detecting

and classifying diseases. Key algorithms cover

Visual Geometry Group(VGG), ResNet, You Only

Look Once Version 3 (YOLOv3), RestNETV2 101,

YOLOv5, Inception-V3, DenseNet, AlexNet,

GoogLeNet, Faster R–CNN, K-most familiar

neighbours, Support vector tool etc. The

performance evaluations related to each model

showed that ResNet has better accuracy and gave

efficient results in differentiating affected and

healthy image patterns with a fully connected layer

using Softmax Function and cross-validation

techniques to improve the potential results.

3 RESEARCH METHODS

This section includes the research methods used in

this paper, including the Study Area, Datasets Used

and Methodologies.

3.1 Study Area

Telangana, a state present in the southern part of

India, spans from latitudes 13’N and 19’N and

longitudes 78’E to 81’E. Telangana has a varied

agricultural terrain, backed by a tropical climate

with warm summers, a moderate monsoon period,

and gentle winters. Farming constitutes the bulk of

Telangana's economy with major crops including

rice, cotton, maize, sorghum, groundnut and

soybean. Rice is the major staple crop sown in large

areas both in the Kharif and Rabi seasons. Climate

and geographical conditions have made Telangana

an important region for agricultural innovation and

development.

Figure 1: Position of Telangana in India (Minhaz, 2023).

3.2 Datasets Used

Data in Climate Resilient Agriculture (DiCRA):

This is a platform with a vast dataset containing

multiple factors contributing to agriculture and key

Predictive Modelling of Agricultural Factors to Maximize Crop Yield

1361

indicators in climate resilience in agriculture.

Environmental features including land surface

temperature (LST), normalized difference vegetation

index (NDVI), temperature, and precipitation as well

as Socio-economic features such as cropland, crop

intensity, etc were used.

Directorate of Economics and Statistics, Ministry

of Agriculture and Farmers Welfare, Government of

India (DESAgri): It contains the historical yearly

yield of all the crops harvested in India. The data is

organized by distinct and the metrics used are

Tonnes/Hectare.

Open Data Telangana: This contains all the

public datasets of Telangana following the open data

policy. The datasets used in this research contain

monthly information regarding temperature (℃),

wind speed (Kmph), humidity (%) and precipitation

(mm) with district and date as the row identifier.

Bhuvan is a web-based platform from the Indian

Space Research Organisation (ISRO) that provides

access to satellite remote sensing data for public use.

We use this to get the Evapotranspiration and

Surface Runoff data.

Evapotranspiration (mm): The data for

evapotranspiration was collected by the National

Remote Sensing Centre (NRSC), one of the agencies

under the National Hydrology Project (NHP). The

evapotranspiration was calculated using the

Modified Preistley Taylor (PT) method. (Ai & Yang,

2016) (Priestley & Taylor, 1972) (Parlange & Katul,

1992). There were a few missing data points due to

technical errors and weather phenomena such as

clouds. Considering the limited meteorological data

availability, using the existing average temperature

data, crop coefficient (Kc) for paddy at different

growth stages and average day length month-wise

were used in the Blaney-Criddle (BC) (French,

1950) equation to fill the missing values. The

Telangana Weather data was combined with the

evapotranspiration data to ensure any dependencies

are captured.

Surface Runoff (mm): The data for surface

runoff was calculated using the Variable infiltration

Capacity (VIC) model, a semi-distributed, physically

based hydrological model, adopted to model water

balance components.

Maplogs day length Dataset: This dataset gives

the day length for the different areas in the world by

using Latitude and Longitude as the key.

Rice Pest Dataset: This rice pest dataset, a subset

of the IP102 dataset, includes of images categorized

into 12 distinct classes that are specific for the

purpose of detecting rice pests. The classes include

rice leaf roller (605 images), rice leaf caterpillar

(475 images), paddy stem maggot (325 images),

Asiatic rice borer (745 images), yellow rice borer

(455 images), rice gall midge (791 images), brown

plant hopper (290 images), rice stem fly (1110

images), rice water weevil (1194 images), rice leaf

hopper (686 images), rice shell pest (480 images),

and thrips (580 images). These were then augmented

with various techniques including vertical flipping,

horizontal flipping, multiplication and linear contrast

adjustment to enhance the dataset.

Rice Leaf Diseases Detection Dataset: The

dataset consists of images showcasing rice leaves in

various conditions, including both healthy and

unhealthy states. This includes healthy rice leaves

(1,085 images), bacterial leaf blight (1,197 images),

brown spot (1,546 images), leaf blast (1,748

images), leaf scald (1,332 images), narrow brown

leaf spot (954 images), neck blast (1,000 images),

rice hispa (1,299 images), and sheath blight (1,629

images). Moreover, the dataset underwent an

augmentation procedure that included the use of

different techniques like rotation, scaling, flipping

etc., to create a larger and more diverse collection of

images.

Rice Insect Pest, Disease Crop Weather

Calendar: Developed by Telangana State

Agricultural University for Nizamabad district is an

all-inclusive guide /tool that informs the occurrence

of insect-pests and diseases at the district level on a

stage-wise basis to take up control measures in time

by thus enabling reduction of losses in yield.

Information regarding the crop, its stages and week-

to-week weather information during the crop season

is essential to forewarn the farmers on

occurrence/prevalence and recommend management

measures against insects, pests and diseases. Farm

operations planned in coordination with weather

information would likely curtail the cost of inputs as

well as other field operations. Rice-insect

pest/disease-weather calendars contain the

favourable conditions required for the occurrence of

key insect pests or diseases and susceptible crop

phenological stages.

3.3 Architecture and Methodologies

The process of creating a predictive analysis model

consisted of the following steps:

i. Yield Prediction: The methodology proposed

to predict Kharif and Rabi crop yields begins by

merging historical crop yield data with the

corresponding weather data based on district and

year, filling in any missing weather values with

column mean. Optimal ranges of monthly

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

1362

temperature, humidity, and rainfall are derived from

growth-stage data for each month to act as

benchmarks.

The preprocessing of the forecasted weather data

creates smoothed features, for instance, 30-day

averages or sums of temperature, rainfall and

humidity. This is then weighted upon in order to

determine how close this data is to falling with its

monthly optimal ranges using a weather scoring

function. The score then combines temperature,

rainfall, and humidity into a single, weighted score

that reflects the overall favourability for crop

growth, adjusted for Kharif and Rabi seasons.

Two models of Random Forests are trained

based on historical weather and yield data to find out

the Kharif and Rabi yields. Yields are predicted in

terms of future weather conditions and adjusted with

the calculated weather score that gives an estimate

of what is the expected yield based upon favourable

or unfavourable conditions. The performance of the

model is validated through R² scores in making sure

predictions are accurate and geared towards district-

level conditions.

ii. Irrigation Calculation: Our approach to

calculating net irrigation required involves three

main models- for the prediction of Precipitation,

Evapotranspiration and Surface Runoff respectively.

Due to limited access to meteorological data,

historical data and weather data were used. Hence

Long Short-Term Models (LSTM) were used to

capture the temporal data instead of standard

regression models with multiple variables used in

previous studies. All the models were trained

district-wise in order to capture the data more

consistently.

Weather Prediction: The Telangana Weather

dataset was used to predict weather features like

Precipitation (mm), Minimum and Maximum

Temperature (℃), Minimum and Maximum Wind

Speed (Kmph), Minimum and Maximum Humidity

(%). To ensure consistency in scaling, these features

were normalized by subtracting the mean and

dividing it by the standard deviation. The data was

divided into specific time steps and an 80-20 split

was used for the train-test division of data. The

architecture used consists of two LSTM layers and

mean squared error was used as the loss function.

Evapotranspiration Prediction: Evapotranspiration

plays a vital role in Irrigation Management. Historical

data calculated by the Modified Preistley Taylor (PT)

(Ai & Yang, 2016) method as recorded by the

Bhuvan dataset and the Blaney-Criddle (BC) equation

(French, 1950) (Choudhary, 2018) were used to

forecast the missing Evapotranspiration values.

The Blaney-Criddle equation is as follows:

ET = (0.0173 Ta – 0.314) Kc *

Ta(D/4465.6) * 25.4

(1)

Where, Ta is mean air temperature (in ℉), Kc is

crop coefficient and D is day length (in hours).

Day length was taken from the Maplogs day

length Dataset and the crop coefficients were

considered as per growth stage of the crop (Mote et

al., 2018).

Feature engineering led us to add four new

features, 1-day lag, 2-day lag, 3-day average and 7-

day average, to the data in order to capture the real-

time changes in the weather patterns. The model

architecture consists of two LSTM layers with

dropout layers to prevent overfitting. The mean

squared error was used as the loss function.

Surface Runoff Prediction: For the prediction of

surface runoff, we considered different models

including Linear Regression, Decision Trees,

Random Forest Regression, Support Vector

Regression (SVR), ensemble models like Gradient

Boosting Regression, XGBoost and LightGBM. The

data was split into training, validation and testing

subsets in a 60-20-20 ratio for each district

separately. We incorporated a custom evaluation

function to identify the best performing model. The

custom evaluation function calculated the mean and

standard deviation for the metrics R

. The best

model was identified by assessing which model had

the highest mean R

while having the least standard

deviation. This was identified to be the Decision

Tree model as shown in Table 3.

The decision tree was trained on the various

weather parameters such as precipitation, humidity,

temperature etc. which were given as the input along

with historical surface runoff data. A 5-fold cross

validation approach, the GridSearchCV technique

from the scikit-learn python library, was used to

finetune the hyperparameters to find the best bias-

variance balance. These hyperparameters were tuned

Figure 2: Surface Runoff Decision Tree for Jayashankar

Bhupalpally district.

Predictive Modelling of Agricultural Factors to Maximize Crop Yield

1363

to maximise the R

and capture the maximum

variance in the data. The best hyperparameters

were max_dept=3, min_samples_leaf=1 and

min_samples_split=2.

To calculate the net irrigation required the

following formula was used:

NI = I - (P -Sr - ETa) (2)

Where, NI is Net Irrigation (mm), I is Ideal

Irrigation (mm), P is Precipitation (mm), Sr is

Surface Runoff (mm) and ETa is Actual

Evapotranspiration (mm).

Net Irrigation was calculated as irrigation

required per hectare of land. So based on the

farmers’ input of farm area the Irrigation required in

total can be calculated.

iii. Preventative Measures Recommendation:

Using the Rice Insect Pest, Disease Crop Weather

Calendar based on the standard week in the year,

crop season and the growth stage of the plant, the

possible pest and disease attacks are indicated. The

probability of them occurring as well as preventative

methods for the pests and diseases are given in a

tabular format.

iv. Pest Detection: The images in the dataset

were resized to be 244x244 pixels to match our base

model ResNet18’s input size. It was then converted

to tensors and normalized using means and standard

deviations. PyTorch’s random split function was

used to slit the dataset into 80-0 train test datasets.

Using transfer learning the base ResNet18 model

was modified by changing the fully connected layer

to fit the number of classes in the dataset. Cross

entropy loss was used to train the model and the

Adam optimiser was used at a learning rate of 0.001.

A progress tracker (tqdm) was used to monitor the

training and validation losses. Early stopping was

employed to ensure the model doesn’t overfit. The

model was evaluated against the training, validation,

and test subsets. Accuracy measures the percentage

of correct predictions compared with true labels. The

model was tested against a final test subset to ensure

generalisation.

v. Plant Disease Detection: ResNet18 is a

commonly used base neural network for many

image-based machine learning applications where

extraction of specific and meaningful features is

required. It allows for easy learning throughout the

network as well as overcoming the degradation

because of the vanishing gradient problem.

Therefore, this model was used in the disease

detection component of our research. We used a

modified and augmented dataset to improve the

quality and classification capability of the model.

For the training and testing stages we utilized the

Pytorch Lightning method and the Tensorflow

preprocess function.

The ResNet18 model consists of 18 layers from

the input to the output layer. Based on the parameter

values it was observed that the ResNet18 model

outperformed the other networks in identifying rice

leaf disease. The model achieved a test accuracy of

77% and test loss of 1.32 when tested on the dataset.

These metrics indicate a fairly efficient model,

although there could be opportunities for additional

optimization to enhance accuracy and decrease loss.

4 EXPERIMENTAL RESULTS

4.1 Crop Yield

The Random Forest model demonstrated strong

performance in predicting crop yields for both

Kharif and Rabi seasons.

Overall, the Random Forest model proved to be a

reliable tool for predicting crop yields in both Kharif

and Rabi seasons. While the model performed

slightly better for Kharif, it still provided accurate

predictions for Rabi as well. The results are

indicated in Table 1.

Table 1: Crop Yield Maximisation Performance Metrics.

Season

Performance Metric

RMSE MAE

Average

Difference

(Predicted –

Actual)

Kharif 0.7525 0.2192 0.1678 0.5335

Rabi 0.7206 0.2319 0.1786 0.5709

4.2 Irrigation Calculation

Irrigation Calculation was split into predictive

analysis of three components- weather prediction,

evapotranspiration forecasting and surface runoff

prediction.

As the evapotranspiration data is combined with

the weather predictive data, it gives a good

indication of the performance of the irrigation

calculation model. The order of measurement is

relatively negligible when compared to

evapotranspiration and hence, not as impactful on

the irrigation model.

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

1364

For Evapotranspiration, prediction was carried

out over the various districts with separate models

for each one. The models were trained for each

district to ensure that the minute changes in weather

and other conditions are captured as accurately as

possible. The mean and standard deviation (SD) of

the performance metrics R

, Accuracy Percentage

and Root Mean Square Error (RMSE) is presented in

Table 2.

The outcome value shows that the model has

fairly taken into consideration the different aspects

relating to the prediction of outcomes - the R² values

range around 0.79 to 0.98 for the different regions

and depict a high value of variance explanation in

most cases. The mean accuracy level of 81%

showcases the effectiveness of the model in all the

geographical locations.

Table 2: Evapotranspiration Performance Metrics.

Statistical

Measure

Performance Metric

Accuracy

RMSE

Mean

0.9035

81.0616

0.2778

0.0448

5.0134

0.0534

In the case of Surface Runoff, our evaluation

focused on the R

and RMSE for each model for all

the districts. We determined the best models by

finding the mean and standard deviation for each

model across the districts.

The results are detailed in Table 3. It was found

that the Decision tree model (DT) had the highest

mean R

value and maintained low RMSE values.

The highest R

was achieved for the district

Khammam with the value of 0.8404.

Table 3: Surface Runoff Models Performance Metrics.

Model

Performance Metric

RMSE

Mean SD Mean SD

Decision Trees 0.7360 0.2056 7.7135 5.7883

Gradient

Boosting

ression

0.5711 0.1457 12.9112 8.3090

LightGBM 0.6867 0.1453 18.9772 14.9264

Linear

ression

0.7129 0.1110 26.3868 14.9429

Random Forest

Regression

0.5430 0.3390 27.7005 29.6897

Support Vector

Regression

0.2888 0.0778 4.3581 5.5488

XGBoost 0.6711 0.1910 14.0355 17.7094

4.3 Anomaly Detection

For anomaly detection the models used were

classification models and hence, the metrics used

were modified as needed. Accuracy along with loss

were used as criteria to measure how well the model

is performing.

Our best performing model on the test dataset for

pest detection showed good generalization as

indicated by a test accuracy of 98.63% and a low test

loss of 0.0437. This implies that the model correctly

classified unseen pest images with precision and

minimum errors, which indicates that the model has

identified the most relevant features without

overfitting to the training data. The close agreement

of test accuracy with training performance confirms

the robustness of the model in pest detection on new

data, thereby reinforcing its appropriateness for

practical deployment in the identification of pests

infesting a crop of rice. The detailed results are

shown in Table 6. The performance metrics are

shown in Table 4.

Table 4: Pest Detection Performance Metrics.

Performance Metric Value

Accuracy 0.9692

Precision 0.9703

Recall 0.9692

F1 Score 0.9690

For disease detection, the model achieved a test

accuracy of 83.96% when tested on the dataset.

These metrics indicate a fairly efficient model,

although there could be opportunities for additional

optimization to enhance accuracy and decrease loss.

The detailed metrics are given in Table 5.

Table 5: Disease Detection Performance Metrics.

Performance Metric (Rice

Disease)

Value

Accuracy 0.8396

Precision 0.8471

Recall 0.8396

F1 Score 0.8414

Predictive Modelling of Agricultural Factors to Maximize Crop Yield

1365

Table 6: Pest Detection Results.

Pest TPR FPR

Rice Leaf Roller 97.69% 0.07%

Rice Leaf Caterpillar 98.53% 0.28%

Paddy Stem Maggot 97.54% 0.43%

Asiatic Rice Borer 97.58% 0.17%

Yellow Rice Borer 98.02% 0.04%

Rice Gall Midge 99.75% 0.48%

Brown Plant Hopper 89.66% 0.11%

Rice Stem Fly 99.46% 1.24%

Rice Water Weevil 98.32% 0.43%

Rice Leaf Hopper 85.01% 0.01%

Rice Shell Pest 97.29% 0.06%

Thrips 97.76% 0.14%

Table 7: Disease Detection Results.

Disease TPR FPR

Bacterial Leaf Blight 89.36% 0.13%

Brown Spot 80.26% 3.16%

Leaf Blast 70.44% 5.23%

Leaf Scald 95.34% 2.57%

Narrow Brown Spot 79.84% 2.27%

Neck Blast 97.20% 0.10%

Rice Hispa 87.56% 1.16%

Sheath Blight 75.69% 0.83%

Tungro 88.39% 0.39%

Healthy 77.24% 2.21%

5 DISCUSSIONS AND

CONCLUSIONS

In crop yield maximisation, the ensemble model

demonstrated strength in both generalizability and

robustness, by minimizing overfitting through a

combination of multiple models. While the proposed

Random Forest model achieves acceptable error

margins for Kharif and Rabi models respectively,

employing similar ensemble techniques as those used

for the benchmark could improve the accuracy even

more. By adding an ensemble technique, the proposed

model could potentially achieve a lower RMSE and

increase resilience to variance, potentially

outperforming the benchmark ensemble model in

terms of robustness and precision for seasonal and

district-level applications. Thus, while the benchmark

model excels in overall error reduction, the proposed

model's tailored approach provides certain practical

benefits in forecasting agricultural yield for localized

and season-specific contexts.

For irrigation calculation, the values for surface

runoff were observed to be several degrees lower

than that of evapotranspiration and precipitation. So,

the effectiveness of our study is indicated by the

performance of our weather and specifically our

evapotranspiration model.

We have also taken into account the lack of data

availability in certain areas by making our model

less dependent on meteorological data, which is

harder to capture. By displaying similar or better

performance than benchmark models, despite having

limited data, our model is more robust in areas

where data is not widely available. This efficiency

along with the high confidence score, makes the

model a good fit for broader applications.

For pest detection, the model used, ResNet 18

can process large amounts of data using data

augmentation and transfer learning, resulting in

higher accuracy using various classifiers. Unlike

traditional methods, deep learning models can

automatically extract relevant features from images

improving accuracy, CNNs are capable of multi-

spectral and hyperspectral processing, enabling the

detection of disease symptoms. Therefore, they can

detect rice diseases accurately and timely. The

results are shown in Table 6 and the performance

metrics are detailed in Table 4.

In case of disease detection, the algorithms use

different data to identify and classify various

diseases, which is important for timely intervention

and disease control. Machine learning models can be

adapted to different locations and environments to

address emerging field conditions. Although this has

been the outcome, there are still some barriers in the

general understanding of the data and the different

machine learning models. Therefore, future study

should be developed more accurately and more

efficient models should be deployed while

improving data collection and documentation.

Hence, using machine learning, agricultural

sustainability can be improved and utilization of

pesticides, fertilizers, and other materials easier and

also reduces the environmental impact besides

enhancing efficiency. The detailed results for

different diseases are shown in Table 7.

The different components of this study come

together to create a comprehensive tool for farmers to

use to ensure that their crops are climate resilient, on a

day-to-day basis. By using predictive analysis, the

effects of climate change can be captured effectively

to provide real-time insights into maximizing crop

yield. Future work includes expanding our models to

other crops as well as to other areas in the country and

beyond. With more data availability a standard, but

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

1366

area-specific, model can be developed for climate

resilient agriculture in India.

ACKNOWLEDGMENT

At the outset, we would like to thank our Professors

at PES University for their unwavering guidance in

ensuring the quality of our research. We are also

extremely indebted to our friends and family for

their help and support.

REFERENCES

Satpathi, A., Setiya, P., Das, B., Nain, A. S., Jha, P. K.,

Singh, S., & Singh, S. (2023). Comparative Analysis

of Statistical and Machine Learning Techniques for

Rice Yield Forecasting for Chhattisgarh, India.

Sustainability, 15(3), 2786. https://doi.org/10.3390/

su15032786

Breiman, L. (2001). Random Forests. Machine Learning,

45(1), 5–32. https://doi.org/10.1023/a:1010933404324

Pang, A., Chang, M. W. L., & Chen, Y. (2022).

Evaluation of Random Forests (RF) for Regional and

Local-Scale Wheat Yield Prediction in Southeast

Australia. Sensors, 22(3), 717. https://doi.org/10.

3390/s22030717

Iqbal, N., Shahzad, M. U., Sherif, E.-S. M., Tariq, M. U.,

Rashid, J., Le, T.-V., & Ghani, A. (2024). Analysis of

Wheat-Yield Prediction Using Machine Learning

Models under Climate Change Scenarios.

Sustainability, 16(16), 6976–6976. https://doi.org/10.

3390/su16166976

Teixeira, R., Cerveira, A., Solteiro, E. J., & Baptista, J.

(2024). Enhancing Weather Forecasting Integrating

LSTM and GA. Applied Sciences, 14(13), 5769–5769.

https://doi.org/10.3390/app14135769

Kadkhodazadeh, M., Valikhan Anaraki, M., Morshed-

Bozorgdel, A., & Farzin, S. (2022). A New

Methodology for Reference Evapotranspiration

Prediction and Uncertainty Analysis under Climate

Change Conditions Based on Machine Learning, Multi

Criteria Decision Making and Monte Carlo Methods.

Sustainability, 14(5), 2601. https://doi.org/10.3390/

su14052601

Granata, F., & Di Nunno, F. (2021). Forecasting

evapotranspiration in different climates using

ensembles of recurrent neural networks. Agricultural

Water Management, 255, 107040. https://doi.org/10.

1016/j.agwat.2021.107040

Yin, J., Deng, Z., Ines, A. V. M., Wu, J., & Rasu, E.

(2020). Forecast of short-term daily reference

evapotranspiration under limited meteorological

variables using a hybrid bi-directional long short-term

memory model (Bi-LSTM). Agricultural Water

Management, 242, 106386. https://doi.org/10.1016/j.

agwat.2020.106386

Gauch, M., Kratzert, F., Klotz, D., Nearing, G., Lin, J., &

Hochreiter, S. (2021). Rainfall–runoff prediction at

multiple timescales with a single Long Short-Term

Memory network. Hydrology and Earth System

Sciences, 25(4), 2045–2062. https://doi.org/10.5194/

hess-25-2045-2021

Minhaz, A. (2023, June 15). Telangana’s erratic summer

showers. Frontline. https://frontline.thehindu.com/the-

nation/agriculture/telangana-erratic-summer-showers/

article66941025.ece?utm_source=relatedstories&utm_

medium=article&utm_campaign=trackRelArt

Ai, Z., & Yang, Y. (2016). Modification and Validation of

Priestley–Taylor Model for Estimating Cotton

Evapotranspiration under Plastic Mulch Condition.

Journal of Hydrometeorology, 17(4), 1281–1293.

https://doi.org/10.1175/jhm-d-15-0151.1

Priestley, C., & Taylor, R. (1972). On the Assessment of

Surface Heat Flux and Evaporation Using

Large-Scale Parameters. Semantic Scholar. https://

doi.org/10.1175/1520-0493(1972)100%3C0081:OTA

OSH%3E2.3.CO;2

Parlange, M. B., & Katul, G. G. (1992). An

advection‐aridity evaporation model. Water Resources

Research, 28(1), 127–132. https://doi.org/10.1029/

91wr02482

French, H. (2021). Determining Water Requirements in

Irrigated Areas From Climatological and Irrigation

Data; TP-96. Hassell Street Press.

Choudhary, D. (2018, March 22). Methods of

Evapotranspiration. https://doi.org/10.13140/RG.2.2.

14533.76007

Mote, K., V. Praveen Rao, Kumar, K., & V. Ramulu.

(2018). Estimation of crop evapotranspiration and crop

coefficients of rice (Oryza sativa L.) under low land

condition. Journal of Agrometeorology, 20(2), 117–

121. https://doi.org/10.54386/jam.v20i2.521.

Predictive Modelling of Agricultural Factors to Maximize Crop Yield

1367