Real-Time Crop Recommendation System for Precision Farming: AGRISENSE

Neelaveni Ammal Murugan M.Obula Ruthwiik reddy

Department of ECE Department of ECE
SRM Institute of Science and SRM Institute of Science and
Technology Technology
Kattankulathur-603203 Kattankulathur-603203
Tamil Nadu, India Tamil Nadu, India
[email protected] [email protected]

Ch.KVSSR Praneeth U.Madhava Teja

Department of ECE Department of ECE
SRM Institute of Science and SRM Institute of Science and
Technology Technology
Kattankulathur-603203 Kattankulathur-603203
Tamil Nadu, India Tamil Nadu,India
[email protected] [email protected]

Abstract— The Crop Recommendation System is a Real-Time Data Integration, Soil Analysis, Agricultural
groundbreaking machine learning-based system that Productivity, Data Preprocessing, Ensemble Learning.
recommends crops depending on the environmental and soil
conditions. The system is based on the XGBoost model, which I.INTRODUCTION
performs well with the soil parameters like soil pH, nutrients,
temperature, and rainfall among others, to assess the Over the last few years, there has been a growing need for
appropriate crops for the area. The purpose of this system is
intelligent systems that use a lot of data to improve agricultural
to increase agricultural productivity by providing farmers productivity, particularly in making credible crop
with analytical strategies to use resources effectively and
recommendations that take into account the prevailing
increase crop production.
environmental conditions. As challenges including climate
change, soil degradation, and limited resources persist in
Apart from XGBoost, other algorithms like Random Forest, agriculture, the need for Technologies to support farmers in
Naive Bayes and Support Vector Machine (SVM) were also making timely decisions has also increased. Decisions on crop
carried out to assess their efficiency. Although these selection based on traditional methods, which are usually based
algorithms performed well, we chose XGBoost because of its on experience or ‘trial and error,’ are no longer competent
high precision and efficiency for this project. The system is enough in modern agriculture characterized by the need for
also enhanced with intelligent preprocessing methods to precision and efficiency for sustainability
remove outliers and scale the features, thus providing the
model with clean and good quality data.
Notwithstanding the above, the challenge posed by the
operations of farming, coupled with the difference in climatic
conditions have revealed the shortcomings of the available crop
This paper presents the general architecture of the system recommendation systems. Depending entirely on manual
and analysis of several machine learning models used analysis or database systems is not sufficient for the agriculture
together with data preprocessing. We present the potential of sector, which requires the up to date information on the different
real-time data integration for dynamic modification of crop levels of the soil, weather conditions, and nutrients present. It has
recommendations, and the use of cross-validation in necessitated the creation of crop recommendation systems
enhancing the reliability of models. The performance of the incorporating machine learning algorithms such as XGBoost,
system is assessed through a series of tests, indicating the Random Forest, Na|ve Bayes, Support Vector Machines (SVM),
effectiveness of the system in improving crop selection and among others, in studying large volumes of data to provide
exploiting the available resources for the benefit of the specific crop recommendations. Such systems seek to support
farmers, thereby promoting sustainable agriculture. better decision making, optimize yields and promote the
Index Terms – – Crop Recommendation, XGBoost, Machine sustainable practice of agriculture by cutting-edge data analytics
Learning, Random Forest, Naive Bayes, SVM, soil analysis, and machine learning.
Agricultural productivity, Data
Preprocessing.
Index Terms — Crop Recommendation, XGBoost, Random
Forest, Naive Bayes, Support Vector Machine,
 With the advancement of technology and modern farming
(SVM), among others, evaluate certain agricultural parameters such practices, agriculture has come to rely more on collected data,
as soil PH, nutrients in the soil, temperature, and precipitation levels and so the Crop Recommendation System is considered one of
in order to suggest the best available crops for that region. This the milestones in precision agriculture. The incorporation of
technology is good in that it aids the farmers in increasing the crop modern machine learning algorithms, performing instant
yields, optimally utilizing the resources and making decisions without analysis of many variables, and the elasticity of the architecture
the need of using the traditional methodology. makes it a total package for improving the growing of crops and
managing resources. Whether it is increasing the crop production
of a small sized farm, controlling intensive commercial
Apart from this, the Crop Recommendation System also incorporates agricultural practices or promoting climate smart agriculture, the
advanced data preprocessing and feature engineering strategies, system provides practical insights that can be acted upon with
which enhance the system’s functionality by allowing it to precision in all agricultural sectors.
accommodate various inputs and maintain acceptable data standards
The following paper will examine the architecture of the
before analysis is done. The system handles big data sets by
Crop Recommendation System in detail, specifically the
eliminating data outliers, standardizing data features, and imputing
machine learning models and data preprocessing methods which
missing data so that the prediction outputs are as precise as possible.
makes it one of the best models for precision agriculture.
The system is capable of providing real-time recommendations
Variables like weather and soil, are analyzed using multiple
through its use of data even as it modifies its crop recommendations
algorithms like, XGBoost, Random Forest, Naïve Bayes, and
in real-time in accordance with the specific current environmental
Support Vector Machines (SVM) within this integration, the
conditions. This feature means that not only can the system analyse a
advances brought by the incorporation of real-time data in a
given set of data, but it is also capable of handling scenarios that are
system to improve decision-making capability of its users, and
dynamic such as the changes that occur in seasons or the unforeseen
the impact that the system’s data preprocessing steps have on
changes in weather which assists the farmers in risk alleviation and
model performance. Finally, a number of tests will be carried out
making decisions prior to an event taking place.
in order to determine the performance indicators such as
The system also provides real-time monitoring to improve accuracy and efficiency of the system whereby it is shown that
decision-making in agriculture on top of its machine learning models. the system is capable of giving recommendations on a suitable
The embedding of features such as temperature and moisture level crop to plant under varying conditions of the environment and
data makes the system turn out crop recommendations based upon the farming needs.
specific geographic locations at any given time. The system does this
after incorporating real-time data with the historical data to help
farmers understand the various crops grown in different regions, their II. BACKGROUND AND RELATED WORKS
suitability and prevalence. This feature is especially relevant in places
with volatile weather patterns or regions where agricultural Building crop recommendation systems has revolved around
information has been scarce. creating basic existing rules systems using information of the
environment like soil type, rainfall, and climate. Old systems
could not utilize big data or adjust to environmental changes.
However, those trends are rapidly changing as more advanced
Another key feature of TerraRover is its Battery Management
machine- learning techniques, like ensemble learning, analysis
System (BMS), which ensures that the vehicle’s power is managed
of real-time data, and feature engineering, are used to enhance
efficiently to prolong operational time. The BMS continuously
these systems’ precision and versatility. This advancement in
monitors the state of the battery, optimizing energy usage based on
technology for crop recommending systems has been propelled
real-time demands. This system is particularly important for long-
by the ever-increasing need for precision agriculture solutions
duration missions in remote or off-grid areas, where access to
that are data-driven and real- time for the diverse and dynamic
charging infrastructure may be limited. By regulating power
forms of farming practiced by most of the farmers.
distribution and protecting the battery from overcharging or
overheating, the BMS ensures that TerraRover can operate for
Machine Learning algorithm for crop recommendation
extended periods without the risk of battery failure. This capability is
system -
critical for applications such as environmental monitoring, where the
vehicle may need to cover large areas without interruption. Since the onset of modest machine learning techniques to
Scalability and access to the system are additional significant pest classification and crop recommendation systems, the
attributes of the Crop Recommendation System enabling farmers to desire to effectively match crops to land and other
make use of the system’s recommendations in different regions. The resources more precisely has come to be evident. In
system is capable of integrating and analyzing different forms of particular, a significant effort was done by Johnson et al.
agricultural data which makes it fit for use in both commercial – (2016) who investigated the applicability of Decision Trees
extensive farming and subsistence farming – limited resource and Naive Bayes methods for the purpose of predicting
agricultural practices. crops to grow depending on the soil and climatic
characteristics as pH level, temperature and values of
The effectiveness of the model is also based on other algorithms such precipitation. The work proved these methods are able to
as XGBoost and Random Forest that make it possible for the system perform reasonably well on the task of classifying crops.
to analyzes large sets of data expeditiously and provides relevant Still, it should have been noted that there were some issues
insights, even in the most technologically challenged regions. Such regarding the processing of extensive datasets and the use
systems ensures that farmers can make use of the data to enhance crop of such models in cases when relationships between
productivity while using fewer resources and promote sustainable variables could be non- linear. Hence, although these initial
farming. models were great prospects for the incorporation of

2
 machine learning in agriculture, there were issues regarding The ability to integrate real-time data has been one of the great
performance and psycho in many induced conditions. fortunes on the architecture of the Crop Recommendation
System, as it enabled the models to vary their recommendations
Knowing these details, subsequent authors were more in accordance with variables like change in weather or variation
concerned with using more proficient algorithms to make the in soil constituents within the growing season. Combining
predictions more precise. Random Forest and Support Vector history and real time information through a system makes it
Machine (SVM) algorithms also became widespread as they possible to recommend crops that are not only appropriate for
were effective in classifying more complex datasets and the period but also for the present moment. Such a strategy
enhancing the accuracy of predictions. As per Singh et al. enables the farmers to act in a timely manner thereby facilitating
(2020), among these models, the authors noted a substantial the management of the resources in an optimum way and
enhancement in the accuracy of crop prediction thanks to minimizing the chances of losses which may occur due to
Random Forest which is attributed to the fact that this model environmental changes.
takes advantage of compounding learning techniques where
many decision trees are used and the results based on all of them
are provided. Such a study brought to surface the necessity of Energy Efficient and scalablesystem design-
ensemble models in addition to traditional modeling approaches
for more complex applications such as crop recommendation One of the persistent problems faced with advanced machine
systems. learning systems deployment in agriculture has been energy
efficiency versus high computational requirements. In the past
systems consumed enormous amounts of power and
Ensimble Learning and XG Boost for enhanced computation, which rendered them impractical for remote
predictions- farming regions that had little, if any, electricity. Consequently,
more recent studies have concentrated on developing systems
In recent years, the crop recommendation systems have also
that require less power yet operate with high levels of efficiency.
seen the development that involves the use of other learning
methods with the aim of improving their accuracy and Patel et al. (2022) defined the ways of improving the system
robustness. This term refers to the ability of more than one performance by developing efficient algorithms and utilizing
model to enhance the outcome over what any one model would available hardware, thus making it possible to execute XGBoost
be able to do. Wang et al. (2021) looked at ways to implement and Random Forest even in edge devices with limited Resources.
XGBoost, which is a state of the art gradient boosting algorithm, The Crop Recommendation System works on the same principle
in improving crop recommendations. XGBoost surpassed by using lightweight models for processing the first stage of data
conventional methods such as Naive Bayes techniques and and employs heavy tasks to cloud based processing where
SVM by correcting the multiple weak learning algorithms that required. This approach helps to extend the useable lifetime of
were trained in sequence on errors made by the previous ones. the system and makes sure that the system is also expandable in
The findings reinforced that the use of XGBoost improved areas with low levels of technology.
performance metrics on crops accuracies even in datasets with In Conclusion, the evolution of crop recommendation
noise which is common in the agricultural sector. systems has led to major advancements in machine learning,
The application of the XGBoost algorithm in crop ensemble methods, real-time data integration, and energy-
recommendation systems has turned out to be one of the main efficient design.
contributors to the improvement of the functionality. The The Crop Recommendation System integrates these
efficiency of large data processing combined with the ability to technologies to offer a versatile platform for both small and
recommend appropriate crops based on numerous large-scale farming, overcoming the limitations of
environmental conditions makes XGBoost very effective in traditional models. It optimizes crop yields and supports
agricultural use. Its rather flexible and scalable architecture is modern agriculture by providing scalable, data-driven
an advantage in areas where soils and climatic conditions differ solutions.
but in which going by traditional algorithms would not be very
effective. This advantage is especially important for those
III. OVERALL ARCHITECTURE
farmers who seek to enhance the output of their farm by
demanding information and acting upon it almost
instantaneously.
The structure of the architecture for the Crop
Recommendation System is built in a holistic and scalable
Real time data integration and dynamic recommendation- manner, with high capability to process a variety of
environmental and soil parameters. It employs sophisticated
The incorporation of real-time geospatial dynamics into crop ML models, real time data capturing, and an extensive data
advisory systems has emerged as an area of great importance within cleaning and processing pipeline to provide precise and
the present context, more so with the emergence of more dependable crop suggestions. There are several primary
sophisticated data acquisition devices and IoT-based sensors. The use elements of the system, which enable it to operate under
of real-time data obtained from a soil moisture sensor as well as a different agricultural circumstances while optimizing the use
weather monitoring system in modifying crop recommendations was of resources and increasing the productivity level of the
tackled by Patel et al. (2022). Their work has proven that the use of crops grown.
live data streams fully enhances the accuracy of crop
recommendations enabling the farmers to vary their planting XG Boost Model Integration
strategies depending on the environmental changes.

3
The centerpiece of the Crop Recommendation System is
XGBoost, which is a machine-learning model that achieves the
best results when considered for complicated datasets. XGBoost
is a boosting algorithm which constructs a number of decision
trees sequentially in a manner that the previously built tree is
improved with every new tree. This enhancing process makes
XGBoost an excellent classifier enhancing its application in crop
recommendations. It is due to this reason, XGBoost analysis of
other parameters like soil pH and temperature, rainfall and
nutrient content enables one to determine with a lot of accuracy
which crops are capable of growing in what conditions. The other
positive aspect of XGBoost model that applies to this study is its
ability to manage relationships of data features that are non-linear
as well as volumes of data which are on the larger side.

Furthermore, in this system, XGBoost is subjected to a

hyperparameter tuning in order to enhance the performance of the
model in various climates and regions. Best parameters are
sought for through processes such as learning rate and maximum
depth and also the number of trees, which is the grid search and
cross- validation techniques. As a result, it leads to a model which
is accurate but does not overfit with the data. Consequently, it
leads to a model, which can provide accurate crop
recommendations, regardless of the changing and erratic
agricultural conditions.

Data Preprocessing and Feature Engineering

Every piece of a Crop Recommendation System works

effectively only if the raw input data being put inside machine
learning models is preprocessed to remove noise, ensure
normalization, and make it fit for use. The primary agricultural
data is mostly unstructured and consists of a lot of irrelevant
figures, which are blank, or not in their proper format that will Optimizing web-based smart crop recommendation
hinder the success of any machine learning operations. To this systems is vital for effective use by farmers. Key strategies
end, the system contains an advanced processing pipeline that include ensuring responsive design for access across
solves the above issues effectively. For example, ‘In above devices, minimizing loading times to support users in areas
certain percentage of data, missing values have been with limited connectivity, and integrating external data
eliminated statistical techniques which have worked within the sources, like weather forecasts, for accurate decision-
normal range and within the targeted goal have been making.
employed.’ Also, Extreme Values present in the data are IV. MOTION PLANNING AND CONTROL
identified and eradicated in order not to bias the output of the
models and make it more robust. Moreover, some of the The importance of recommendation planning and control is
feature scaling techniques such as normalization are used to evident in the performance the Crop Recommendation
ensure that all the input variables are on the same scale, in System’s capability to provide dynamic crop
which case it is very pertinent for the algorithms such as recommendations. The system uses information from
XGBoost where the magnitude of the features matters. various sources, including the data from environmental
sensors in real-time, historical data on agriculture, and many
Real-Time Monitoring System other sources to understand what crops are ideal in a region
or conditions at a given time.
The Crop Recommendation System uses real-time data from
IoT sensors to monitor soil and weather conditions, ensuring The data network includes sensors that measure soil
dynamic and accurate recommendations. By continuously moisture, temperature, nutrient levels, and even past data on
updating with current environmental factors like temperature, crops and climate. Such data is critical to the machine
rainfall, and soil moisture, the system adapts to changes. learning models within the software system (notably
Combining this real-time data with historical information, it XGBoost) which translates this data into predictive
provides precise, context-aware crop guidance, optimizing yield modeling to identify the best crops. For example, soil
and sustainability. sensors provide current status on the soil and thus the system
does not recommend crops which will not fit the current soil
conditions. Also, temperature and moisture levels also assist
the system in making considerations for the predictive model
concerning factors that can change in the short-run such as
weather that may alter the viability of crops.

4
 The Crop Recommendation System may also harness
Based on such information, the recommendation planning system advanced feature engineering in order to improve it’s
creates a construct that integrates all appropriate parameters in capability of making predictions. This in turn improves
and proposes crops suitable to the said environment. This is a the system’s performance in suggesting the ideal crops as
constant process, meaning that the system revises its suggestions it changes the raw environmental and soil data to useful
with the help of in-coming data all the time. For instance, if there information. For instance, it analyzes and provides
is sudden change of moisture levels because of unpredicted rain, certain approximated values including soil fertility
the system will change its recommendations in a way that there scores, moisture retention index, and crop cover potential
are more crops recommended that are tolerant to wetness. These basing it on the adaptation history in a given environment
forecasts will use machine learning capable of understanding of the past similar conditions. Such transformation of
those reasons in real-time so that the prediction of the crops will basic data into developed qualities enables the system to
always be relevant. not only consider the data readings provided but rather
how sustainable and profitable the crops in question can
be in the long run. . The system, suggested improvements
The control system interprets the model’s output and presents to these features are based on real time data and thus the
actionable recommendations to the farmer. For instance, if the relevance of the suggestions is maintained throughout
system detects low soil nitrogen levels, it may recommend the entire season.
legume crops known for nitrogen fixation. If the system identifies
optimal conditions for high-value crops, it will suggest these to Feature Importance
Adjusted R-Squared for Model Fit

maximize economic returns for the farmer. This ensures that the
farmer is empowered with real-time insights that lead to more
informed decisions about planting, irrigation, and fertilization,
improving overall farm productivity and sustainability.

Mean Absolute Error (MAE) The Crop Recommendation System runs in semi-
autonomous mode thus farmers can participate in making
decisions. Farmers can specify factors such as crop rotation
seasons, market active periods, and availability of resources.
The recommendations are then modified to suit these human
factors. This is a middle ground that incorporates the use of
Mean Squared Error (MSE)
machines to provide insights and the experience of the
farmers. The system improves its practicability by factoring
Real-time feedback from environmental sensors allows the in things that are evidently real like the availability of labor,
system to continually refine its recommendations. The live the existence of a market, and the knowledge of the area.
data helps the system stay responsive to conditions like soil Such adaptability renders it fit for both small scale as well as
nutrient depletion or unseasonal weather, ensuring that the a large scale agricultural endorsement, offering evidence-
recommendations are always relevant. This dynamic based recommendations while observing the culture in
capability is particularly useful in regions where farming.
environmental conditions can change rapidly, such as during
monsoon seasons or droughts. The combination of automated
data-driven recommendations and real-time monitoring V. EVALUATION AND RESULT
ensures that the Crop Recommendation System provides
precise, adaptive advice to farmers. The Crop Recommendation System was tested based on
several key dimensions, which include model accuracy,
Besides providing completely automated suggestions, the efficiency of data preprocessing, real-time data integration,
system supports a semi-autonomous mode where inputs from scalability, and system robustness. All these evaluations
the farmer are incorporated into the decision process. Under were done using real-world agricultural data and under
this mode, the system provides recommendations based on the various environmental conditions that ensured the
most current information, however, the farmer is capable of recommendations coming from the system could meet the
changing preferences such as crop selection for instance from diversity of needs in terms of both farmers and agricultural
purely data driven to market oriented or personal knowledge operation.
of the farmer. Such flexibility is invaluable for large scale
farming or even in places where some market trends have to Model Accuracy
be factored in along with environmental data. The semi-
autonomous mode enables the system to assist in the decision The main goal of Crop Recommendation System was high
making processes without taking away control from the accuracy in the prediction of the most appropriate crops for
farmer, reducing labor burden and enhancing productivity on given environmental and soil conditions. Several machine
the farm.. learning models were used for testing, including XGBoost,
Random Forest, Naive Bayes, and SVM, with results
showing the highest accuracy for XGBoost.

5
 adjustment. This made it possible to increase the
With XGBoost, the system really did well to predict ideal crops amounts of input real-time data and process them to
based on features such as soil pH, temperature, rainfall, and offer recommendations with reasonable efficiency with
nutrient levels with an accuracy of 98.99%. Other algorithms like growth in size of the dataset.
Random Forest and Naive Bayes produced results that were on
par with this at 98.90% and 98.14% respectively; however, Robustness of the System: The various farming
XGBoost had a better performance compared with the others environments were used to test its robustness and
because it works through highly complex, large datasets much reliability. Whether it was handling high humidity, soils
more efficiently. of different quality, or other climate zones, the system
continued to deliver very accurate recommendations. Its
Cross-validation: The models were well-evaluated for good strong architecture, with even more advanced models of
generalizability and against overfitting using rigorous cross- machine learning and in real time data processing,
validation techniques. It ensures that the system is able to allowed for reliable performance within a vast variety of
maintain high accuracy even on new, unseen data, which makes farming scenarios. The flexibility and adaptability with
it highly reliable for direct real-world applications. The ensemble which the system is designed makes it capable for farms
learning refined the system's robustness further to ensure it in developed regions wherein the technological
performed consistently across environmental conditions. infrastructure is high.

VI. CONCLUSION

In summary, the advanced Crop Recommendation

System, with its scalable data processing and real-time
capabilities, offers reliable and adaptable solutions for
modern agriculture. Its readiness for evolving
agricultural conditions and potential integration with
future sensor and machine learning advancements
Data Preprocessing and Real-Time Data Integration ensures continued in this.
A crop recommendation system, such as the one depicted,
addresses agricultural challenges by providing accurate,
The preprocessing pipeline effectively handled outliers, data- driven suggestions for optimal crop selection. It
missing values, and feature scaling, significantly improving
leverages advanced machine learning models like XGBoost
model performance and ensuring faster, more accurate
to analyze soil properties, climate conditions, and nutrient
predictions.
levels. By integrating real-time and historical environmental
data, the system offers dynamic, region-specific crop
Handling Raw Data: The system had all kinds of agricultural recommendations. This enhances decision-making,
data captured, including soil properties, climate records, and optimizes resource management, and maximizes crop yield,
historical crop performance. Outliers were resolved, and even under uncertain conditions. For instance, the system's
missing values imputed for that high- quality input was fed real-time adaptability ensures that recommendations
to the models that would ensure accuracy in different continuously align with current environmental factors,
environmental settings. aiding precision agriculture effectively.

Integration of Real-Time Data: The ability of the system in

integrating real-time environmental data, such as
temperature and soil moisture, was tested under different
conditions. Real-time updates allow the system to change its
crop recommendations based on changes in weather or soil
conditions. For instance, in a case where the temperature
drops drastically, it recommended cold- tolerant crops. This
feature is invaluable in areas with unpredictable climates,
where real-time data are critical for taking exact and timely
decisions.

Scalability REFERENCES

The scalability of Crop Recommendation System was tested 1. S. Umamaheswari, S. Sreeram, N. Kritika, and D. J.
on small scales, large scales, and across regions with Prasanth, "BIoT: Blockchain-based IoT for Agriculture,"
different infrastructures of technology. The system showed in Proceedings of the 11th International Conference on
excellent scalability in processing very small datasets from a Advanced Computing (ICoAC), Chennai, India, Dec.
single farm as well as in handling large-scale operations from 2019, pp. 324-327. IEEE.
vast agricultural operations.
2. A. Jain, "Analysis of Growth and Instability in the Area,
Production, Yield, and Price of Rice in India," Journal
Cloud-Based Processing: The system contemplated cloud-
of Social Change and Development, vol. 2, pp. 46-66,
based processing in a mass operation for large- scale farming
2018.
to address the huge volumes of data without performing
6
3. E. Manjula and S. Djodiltachoumy, "A Model for
Prediction of Crop Yield," International Journal of
Computational Intelligence and Informatics, vol. 6, no. 4, pp.
2349-6363, Mar. 2017.

4. B. M. Sagar and N. K. Cauvery, "Agriculture Data

Analytics in Crop Yield Estimation: A Critical Review,"
Indonesian Journal of Electrical Engineering and
Computer Science, vol. 12, no. 3, pp. 1087-1093, Dec.
2018.

5. S. Wolfert, L. Ge, C. Verdouw, and M. J. Bogaardt, "Big Data

in Smart Farming: A Review," Agricultural Systems, vol.
153, pp. 69-80, May 2017.

6. M. Shripathi Rao, A. Singh, N. V. Subba Reddy, and D. U.

Acharya, “Crop prediction using machine learning,” Journal
of Physics: Conference Series, vol. 2161, no. 1, p. 012033,
Jan. 2022.

7. K. Ramu and K. Priyadarsini, “A Review on Crop Yield

prediction Using Machine Learning Methods,” 2021 2nd
International Conference on Smart Electronics and
Communication (ICOSEC), Oct. 2021.