A SURVEY ON A MODEL FOR PESTICIDE RECOMMENDATION USING MACHINE LEARNING

Computer Science & Engineering: An International Journal (CSEIJ), Vol 15, No 1, February 2025
DOI:10.5121/cseij.2025.15104 27
A SURVEY ON A MODEL FOR PESTICIDE
RECOMMENDATION USING
MACHINE LEARNING
Vijaya lakshmi S Abbigeri, Geetha D Devanagavi
School of Computing and Information Technology, REVA University,
Bengalore, India
ABSTRACT
Pesticides are necessary to ensure the security ofthe world’s food supply by boosting
agricultural productivityand crop yields. Nonetheless, the excessive and improper use of
these chemicals poses grave threats to human health, wildlife,and the fragile ecological
balance. The development of accurate and efficient pesticide recommendation systems is
vital for mitigating these environmental and health-related risks while sustaining necessary
agricultural productivity. This survey paper provides a comprehensive summary of the
several machine learning algorithms that have been applied for the purpose of pesticide
recommendation, highlighting their capabilities, limitations, and possible directions for
future study and development in this criticalfield.
KEYWORDS
Pesticide Recommendation, agricultural productivity, machine learning
1. INTRODUCTION
Utilizing pesticides has become an integral component of contemporary farming methods,
facilitating farmers tomitigate the adverse impacts of pests, weeds, and diseases oncrop yields [1].
While the judicious application of these chemicals has contributed to the remarkable increase in
agricultural productivity over the past few decades, the indiscriminate and excessive use of
pesticides has led to a host of environmental and health-related concerns. The presence of
pesticide residues in food, contamination of water bodies, and the detrimental effects on non-
target organisms, including beneficial insects and wildlife, are well-documented consequences of
unsustainable pest management practices [2].
Researchers have looked into the possibilities of machine learning approaches to create more
effective and focused pesticide recommendation systems in response to these difficulties. These
models aim to provide farmers with precise guidance on the appropriate pesticide selection,
dosage, and application timing, thereby reducing the overall reliance on pesticides while
maintaining crop productivity. By utilizing machine learning’s potential, large volumes of data
from multiple sources can be analyzed by these systems, such as field conditions, weather
patterns, and historical pest management records, to deliver customized recommendations that
optimize pesticide use and minimize environmental impact. This survey paper delves into the
various machine learning models that have been used to prescribe pesticides, highlighting their
capabilities, limitations, and possible avenues for further investigation in this critical field.

28
2. MACHINE LEARNING APPROACHES FOR PESTICIDE
RECOMMENDATION
In the last ten years, academics have investigated numerous machine learning methodologies for
creating pesticide recommendation systems. One method involves the utilization of deep learning
algorithms, which have demonstrated remarkable performance in tasks such as image recognition
and natural language processing [3] [4] [5]. These algorithms can be trained on large datasets of
crop imagery and pest identification information to accurately detect and classify various pests,
enabling targeted and timely pesticide application. The below fig.1.shows pesticide
recommendation system.
2.1. Supervised Machine Learning (ML Models for Pesticide Recommendation
Classification Models:
These models are utilized when the objective is to predict a categorical output, for example the
specific type of pesticide to recommend. In these models, the input features can include factors
like crop type, pest characteristics, environmental conditions, and historical pest management
data, the output is the recommended pesticide [6][7].
 Decision Trees:
[7] These models are relatively straightforward to interpret and can accommodate both
categorical and numerical input variables. These models are useful for clarifying the main
elements influencing recommendations for pesticides.
 Random Forests:
[7] Random forests integrate multiple decision trees to enhance predictive performance and
mitigate overfitting. They are robust and capable of handling high- dimensional data effectively.
Support Vector Machines: [8] SVMs have proven to be effective in classifying complex datasets,
especially when the connections among the input feature relationships exhibit non-linear patterns.
 Naive Bayes:
[7] The Naive Bayes model is straight forward and computationally efficient, making it well-
suited for large-scale datasets. However, it relies on the assumption of feature independence,
which could pose a potential limitation in certain scenarios where the relationships between the
input variables are more complex.
 K-Nearest Neighbors:
[7] The K-Nearest Neighbors model operates by categorizing fresh data samples depending on
their resemblance to previously categorized observations. This approach is straightforward to
implement, but it can be computationally intensive for large-scale datasets.

29
•
•
Fig. 1. Pesticide recommendation system.
Regression Models:
These models are utilized when the objective is to predict a continuous output, such as the
optimal dosage of a pesticide. They aim to establish a functional relationship between the input
features and the target variable, which in this case would be the recommended pesticide
application rate.
Linear Regression: The target variable and the input features are assumed to have a simple linear
relationship by this linear regression model. Its ease of usage permits for simple comprehension
and interpretation, but it may not be well- suited for addressing more complex, non-linear
relationships within the data [9].
Artificial Neural Networks: These models can learn complex, non-linear relationships from
data, making them appropriate for handling intricate patterns in pesticide recommendations. They
can be computationally expensive to train[10].
2.2. Unsupervised Machine Learning (ML Models for Pesticide Recommendation)
Clustering: This helps group similar data points, which in this context could be:
Field Segmentation: Cluster fields with similar soil properties, pest pressures, or historical
treatments. This makes more focused recommendations than a one-size-fits-all approach [11].
Pest Identification: Cluster images or sensor data to identify distinct pest types or infestation
patterns, even without labeled training data [12].
Treatment Response Grouping: Cluster fields based on how well they responded to past
pesticide applications, potentially revealing hidden factors influencing effectiveness [12].
Dimensionality Reduction: Agricultural datasets can be huge! These techniques help simplify
the data while preserving important information:
Principal Component Analysis: Finds the most important combinations of features (soil
nutrients, weather, etc.) that explain most of the variation in pesticide needs. This can simplify
models and improve their performance[13].

30
t-SNE: Useful for visualizing high-dimensional data. Could help experts visually identify clusters
of fields with un-usual pesticide requirements, prompting further investigation.
Association Rule Mining: This uncovers "if then" rules within data. For example:
Co-occurring Pests: "If pest A is detected, there’s a high likelihood of also finding pest B." This
informs broader treatment strategies.
Treatment Patterns: "Farmers who successfully con- trolled pest X in the past often used
pesticide Y early in the season." This provides insights for recommendations.
3. DIFFICULTIES AND PROSPECTS FOR THE FUTURE
Although machine learning models show potential for improving pesticide recommendation
systems, there are stilla number of significant obstacles to overcome. Combining various data
sources, like genomics, environmental, weather, soil, and field management records, can improve
the predictive power of these models and lead to more holistic and effective recommendations.
By incorporating a large variety of relevant data, these models are more able to depict the
complex relationships between different components that influence optimal pesticide selection
and application [14][15].
Several case studies showcase the potential of machine learning for pesticide recommendation:
The below table in Fig.2 shows case studies of different machine learning for pesticide
recommendation.
Study Approach Results
Predicting
Cotton
Bollworm
Infestation
using Deep
Learning
Convolutional
Neural
Networks
Accurate prediction of
bollworm infestation
based on image data,
enabling timely
intervention and
reducing pesticide use.
Optimizing
Pesticide
Application
for Rice Blast
Disease
Reinforcement
Learning
Improved rice blast
control with reduced
pesticide application,
minimizing
environmental impact
and cost.
Fig. 2. case studies of different machine learning for pesticide recommendation.
Improving the interpretability of these models, particularly the "black-box" approaches like neural
networks, is crucial for gaining the trust of farmers and facilitating their adoption. Developing
more transparent and explainable models can help farmers understand the reasoning behind the
recommendations, which is essential for their acceptance and implementation.
The development of integrated decision support systems that integrate machine learning models
with expert knowledge and practical experiences can result in more comprehensive and
practically viable solutions. By integrating the strengths of both data-driven and expertise-based
approaches, these systems can provide customized recommendations that account for the nuances
of local conditions and farming practices [16][17].

31
Investing in long-term data collection and monitoring efforts to build comprehensive datasets is
necessary for developing dependable and accurate machine learning models for pesticide
recommendations. Comprehensive and high-quality data is the basis for creating prediction
models that can effectively address the complex challenges in sustainable pest management [11].
Finally, the incorporation of machine learning methods with a multifaceted understanding of crop
production and protection
holds significant potential to revolutionize pesticide recommendation systems. By optimizing
pesticide use, these systems can contribute to more sustainable and environmentally-friendly
agriculture, ultimately benefiting both farmers and the broader ecosystem [18][19].
5. CONCLUSION
The development of innovative machine learning models has appeared as a viable strategy to
improve crop productivity while mitigating the environmental and dangers to one's health from
overusing pesticides. Supervised learning algorithms, such as decision trees, random forests, and
artificial neural networks (ANN), have shown the capability to accurately predict appropriate
pesticide types and dosages based on various agronomic, environmental, and pest-related factors.
Additionally, unsupervised learning methods, including clustering and dimensionality reduction
techniques, can uncover hidden patterns and relationships within complex agricultural datasets,
enabling more nuanced and tailored recommendations.
However, the successful implementation of these machine learning-based pesticide
recommendation systems requires addressing several key challenges. Integrating diverse data
sources, improving model interpretability, and developing comprehensive systems for making
decisions are crucial steps towards creating practical and trustworthy solutions for farmers. By
addressing these challenges, incorporating machine learning into pesticide recommendation
systems can contribute to more sustainable and environmentally-friendlyagricultural practices,
ultimately benefiting both farmers and the wider ecosystem.
REFERENCES
[1] N. Alexandridis, G. Marion, R. Chaplin‐Kramer, M. Dainese, J. Ekroos, H. Grab, M. Jonsson, D. S.
Karp, C. Meyer, M. E. O’Rourke, M. Pontarp, K. Poveda, R. Seppelt, H. G. Smith, E. A. Martin,
and Y. Clough, "Models of natural pest control: Towards predictions across agricultural landscapes,"
Biological Control, vol. 163, p. 104761, Nov. 2021. [Online]. Available:
https://doi.org/10.1016/j.biocontrol.2021.104761
[2] A. Sharma, A. Shukla, K. Attri, M. Kumar, P. Kumar, A. Suttee, G. Singh, R. P. Barnwal, and N.
Singla, "Global trends in pesticides: A looming threat and viable alternatives," Ecotoxicology and
Environmental Safety, vol. 201, p. 110812, Sep. 1, 2020. doi: 10.1016/j.ecoenv.2020.110812.
[3] M. Chithambarathanu and M. K. Jeyakumar, "Survey on crop pest detection using deep learning and
machine learning approaches," Journal of Visual Communication and Image Representation, vol. 82,
no. 27, pp. 42277-42310, Apr. 11, 2023. doi: 10.1007/s11042-023-15221-3.
[4] X. Cheng, Y. Zhang, Y. Chen, Y. Wu, and Y. Yue, "Pest identification via deep residual learning in
complex background," Computers and Electronics in Agriculture, vol. 141, pp. 351-356, Sep. 1,
2017. doi: 10.1016/j.compag.2017.08.005.
[5] A. Kamilaris and F. X. Prenafeta Boldú, "Deep learning in agriculture: A survey," Computers and
Electronics in Agriculture, vol. 147, pp. 70-90, Feb. 22, 2018. doi: 10.1016/j.compag.2018.02.016.
[6] A. Kumar, S. Sarkar, and C. Pradhan, "Recommendation system for crop identification and pest
control technique in agriculture," Proceedings of the International Conference on Communication
Systems and Signal Processing, Apr. 1, 2019. doi: 10.1109/iccsp.2019.8698099.
[7] S. Pudumalar, E. Ramanujam, R. H. Rajashree, C. Kavya, T. Kiruthika, and J. R. Nisha, "Crop

32
recommendation system for precision agriculture," Proceedings of the International Conference on
Advances in Computing, Communication and Automation, Jan. 1, 2017. doi:
10.1109/icoac.2017.7951740.
[8] M. W. H. Wang, J. M. Goodman, and T. E. H. Allen, "Machine learning in predictive toxicology:
Recent applications and future directions for classification models," Chemical Research in
Toxicology, vol. 34, no. 2, pp. 217-239, Dec. 23, 2020. doi: 10.1021/acs.chemrestox.0c00316.
[9] A. Schneider, G. Hommel, and M. Blettner, "Linear regression analysis," DeutschesÄrzteblatt
International, vol. 107, no. 44, pp. 776-782, Nov. 5, 2010. doi: 10.3238/arztebl.2010.0776.
[10] K. P. Ferentinos, C. P. Yialouris, P. Blouchos, G. Moschopoulou, V. Tsourou, and S. Kintzios, "The
use of artificial neural networks as a component of a cell-based biosensor device for the detection of
pesticides," Procedia Engineering, vol. 47, pp. 989-992, Jan. 1, 2012. doi:
10.1016/j.proeng.2012.09.313.
[11] R. H. L. Ip, L. Ang, K. P. Seng, J. Broster, and J. Pratley, "Big data and machine learning for crop
protection," Computers and Electronics in Agriculture, vol. 151, pp. 376-383, Aug. 1, 2018. doi:
10.1016/j.compag.2018.06.008.
[12] J. G. A. Barbedo, "Detecting and classifying pests in crops using proximal images and machine
learning: A review," Artificial Intelligence, vol. 1, no. 2, pp. 312-328, Jun. 24, 2020. doi:
10.3390/ai1020021.
[13] S. Mishra, D. Mishra, S. Das, and A. K. Rath, "Feature reduction using principal component
analysis for agricultural data set," Proceedings of the International Conference on Electronics and
Communication Technology, Apr. 1, 2011. doi: 10.1109/icectech.2011.5941686.
[14] D. Elavarasan, P. M. D. R. Vincent, V. Sharma, A. Y. Zomaya, and K. Srinivasan, "Forecasting
yield by integrating agrarian factors and machine learning models: A survey," Computers and
Electronics in Agriculture, vol. 155, pp. 257-282, Dec. 2018. [Online]. Available:
https://doi.org/10.1016/j.compag.2018.10.024
[15] M. S. H. Talukder, R. B. Sulaiman, M. R. Chowdhury, M. S. Nipun, and B. H. Hassine,
"PotatoPestNet: A CTInceptionV3-RS-based neural network for accurate identification of potato
pests," arXiv preprint, Jan. 1, 2023. doi: 10.48550/arxiv.2306.06206.
[16] Indu, A. S. Baghel, A. Bhardwaj, and W. Ibrahim, "Optimization of Pesticides Spray on Crops in
Agriculture using Machine Learning," Hindawi Publishing Corporation, vol. 2022, pp. 1-10, Sep.
2022. [Online]. Available: https://doi.org/10.1155/2022/9408535
[17] S. Sharma, A. Partap, M. A. D. L. Balaguer, S. Malvar, and R. Chandra, "DeepG2P: Fusing Multi-
Modal Data to Improve Crop Production," Cornell University, Jan. 2022. [Online]. Available:
https://doi.org/10.48550/arxiv.2211.05986
[18] J. A. Rosenheim, B. N. Cass, H. M. Kahl, and K. P. Steinmann, "Variation in pesticide use across
crops in California agriculture: Economic and ecological drivers," Science of The Total
Environment, vol. 733, p. 138683, Sep. 1, 2020. doi: 10.1016/j.scitotenv.2020.138683.
[19] B. T. Bestelmeyer, G. S. Marcillo, S. E. McCord, S. B. Mirsky, G. E. Moglen, L. Neven, D. P. C.
Peters, D. C. S. Djebou, and T. Wakie, "Scaling Up Agricultural Research With Artificial
Intelligence," IEEE Computer Society, vol. 22, no. 3, pp. 33-38, May 2020. [Online]. Available:
https://doi.org/10.1109/mitp.2020.2986062

A SURVEY ON A MODEL FOR PESTICIDE RECOMMENDATION USING MACHINE LEARNING

More Related Content

Similar to A SURVEY ON A MODEL FOR PESTICIDE RECOMMENDATION USING MACHINE LEARNING (20)

More from CSEIJJournal (20)

Recently uploaded (20)

A SURVEY ON A MODEL FOR PESTICIDE RECOMMENDATION USING MACHINE LEARNING