论文信息
论文解读
AI 辅助整理- 方法
- 采用ANN等机器学习模型,基于环境、土壤、水分和气象参数建模对比。
- 发现
- ANN预测精度最优(RMSE 3136.8、MAE 2502.2、R² 0.073),优于传统方法。
- 机会点
- R²极低,说明特征与模型仍不足,可引入遥感、时序气象与深度学习提升预测稳健性。
论文摘要
基于环境、土壤、水分和气象参数的作物产量预测已成为一个重要的研究领域,回应了可持续农业实践和粮食安全日益增长的需求。本研究采用机器学习方法,优化阿达马瓦州米奇卡地方政府区农业景观中的作物产量生产。利用广泛的作物数据集,运用机器学习技术对影响作物产量的关键因素和模式进行分析、解释和揭示。主要目标是开发一个稳健的预测模型,为农民和利益相关者提供可操作的见解,以提高农业生产力。结果表明,与传统方法相比,基于机器学习的方法在产量预测准确性方面有显著提升。人工神经网络(ANN)具有最低的均方根误差(RMSE,3136.8)、最低的平均绝对误差(MAE,2502.2)和较高的决定系数(R²,0.073255),表明其预测最为准确。这些发现凸显了人工智能在推进精准农业、实现资源高效型耕作和增强粮食安全方面的变革潜力。本研究还指出了未来研究的方向,包括优化资源配置策略、识别具有抗逆性的作物品种、预测和减轻作物病害,以及绘制不同作物的土壤适宜性图。这些努力将进一步推动智能农业系统的采用,在提高生产力和可持续性的同时,支持向气候适应型耕作实践的转型。
全文
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE AND MATHEMATICAL THEORY E-ISSNInternational Journal of Computer Science and Mathematical Theory (IJCSMT) E -ISSN 2545-5699
P-ISSN 2695-1924 Vol 10. No.6 2024 www.iiardjournals.org
IIARD – International Institute of Academic Research and Development
Page 119
Machine Learning Approach Leveraging Crop Datasets for
Optimizing Crop Yield Production
1Bulus Bali, 2Zakawa N. Ngida, 1Isacha Habila
1Department of Computer, Adamawa State University, Mubi Adamawa State, Nigeria.
2Department of Botany, Adamawa State University, Mubi Adamawa State, Nigeria.
bali930@asdu.ng
DOI: 10.56201/ijcsmt.v10.no6.2024.pg119.132
Abstract
Crop yield prediction based on environmental, soil, water, and weather parameters has become a
vital area of research, addressing the growing need for sustainable agricultural practices and
food security. This study adopts a machine learning approach to optimize crop yield production
within the agricultural landscape of Michika Local Government Area, Adamawa State. Applying
extensive crop datasets, machine learning techniques are utilized to analyze, interpret, and
uncover critical factors and patterns influencing crop yields. The main objective is to develop a
robust predictive model that empowers farmers and stakeholders with actionable insights, to
enhance agricultural productivity. The results demonstrate a substantial improvement in yield
predicting accuracy through machine learning -based methods compared to tr aditional
approaches. ANN with the l owest RMSE (3136.8), the lowest MAE (2502.2), and higher R²
(0.073255), indicates the most accurate predictions. These findings underscore the transformative
potential of artificial intelligence in advancing precision agriculture, enabling resource -efficient
farming, and bolstering food security. This study also highlights avenues for future research,
including optimizing resource allocation strategies, identifying resilient crop varieties, predicting
and mitigating crop diseases, and mapping soil suitability for diverse crops. Such efforts would
further drive the adoption of smart agricultural systems, enhancing productivity and sustainability
while supporting the transition to climate-resilient farming practices.
Keywords: Crop datasets; Machine Learning; Predictive modeling; Crop yield; Weather
parameters
- Introduction
Modern agriculture faces significant challenges, including meeting the rising demand for food
production, addressing resource constraints, and ensuring environmental sustainability. As the
population of Michika expands, the agricultural sector encounters multifaceted challenges such as
climate change, soil degradation, and suboptimal resource utilization. Traditionally, farmers have
primarily utilized conventional farming methods, whi ch, unfortunately, lack precision, resulting
in reduced productivity and prolonged time consumption. Precision farming, leveraging advanced
technologies, enhances productivity by precisely identifying the essential practices for each season
(Durai, & Shamili, 2022). Integrating machine learning techniques presents promising avenues for
tackling these challenges by optimizing agricultural practices and elevating productivity. AI -
driven technologies provide innovative solutions to refine agricultural processes, enhance yield
International Journal of Computer Science and Mathematical Theory (IJCSMT) E -ISSN 2545-5699
P-ISSN 2695-1924 Vol 10. No.6 2024 www.iiardjournals.org
IIARD – International Institute of Academic Research and Development
Page 120
predictions, optimize resource management, and mitigate the adverse effects of climate change on
crop production. This study addresses the need to utilize machine learning algorithms in the
agricultural domain, specifically focusing on their empirical effectiveness in optimizing crop yield
production.
The agricultural sector in Michika, like other parts of Nigeria, faces challenges such as
unpredictable weather conditions, inadequate knowledge of soil nutrient levels, pest and disease
management, and inefficient resource use. These challenges contribute to low productivity and
food insecurity. Conventional farming methods fail to address the complexities of modern
agricultural systems, leaving farmers vulnerable to losses and inefficiencies. Accurate crop yield
prediction remains a significant hurdle, as it requires integrating multiple factors, such as climate
conditions, soil characteristics, and farming practices (Prabavathi & Chelliah, 2022). Addressing
this gap demands advanced tools and methodologies that can provide actionable insights and
empower farmers to make informed decisions (Mgendi, 2024).
This study aims to leverage machine learning techniques to optimize crop yield in Michika by
developing predictive models that integrate environmental, soil, and management factors.
Specifically, the objectives are to employ machine learning algorithms on historical data for crop
yield prediction, compare various machine learning techniques for forecasting crop yields, and
assess the effectiveness of these techniques in predicting yields to recommend the most suitable
approach tailored to Michika's conditions.
There is a pressing need to enhance agricultural productivity in Michika Local Government Area
Adamawa State, like other parts of Nigeria. Agriculture plays a crucial role in the economy of the
region, and optimizing crop yield production can contribute to food security, poverty reduction,
and overall economic development (Jiya et al., 2023) . Leveraging machine learning techniques
will harness the power of algorithms to analyze large agricultural datasets and uncover patterns
that can significantly enhance crop yield production. Furthermore, the utilization of crop datasets
that contain information related to climate conditions, soil quality, farming techniques, and crop
performance, among others offers an opportunity to gather valuable insights into agricultural
practices specific to the area. Leveraging machine learning algorithms for optimizing crop yield
production based on historical data in the area holds great potential to address agricultural
challenges, enhance food security, and contribute to the socio-economic development of the region
2. Related works
Crop yield prediction is a pivotal factor in agricultural success and efficiency, yet it poses
significant challenges to farmers. Knowledge gaps regarding surplus harvests, unpredictable
weather, seasonal rainfall, soil nutrients, fertilizer accessibility, pest management, and post-harvest
losses contribute to declining crop production (Abdulraheem et al., 2022 ). Addressing these
challenges requires a holistic approach that integrates advanced technologies and disseminates
knowledge to empower farmers (Mgendi, 2024) . In the quest to optimize crop yield, farmers
require timely and insightful guidance for predicting productivity, conducting analyses, and
unlocking the full potential of their crops. The real challenge lies in the accuracy of yield
prediction, where tapping into a farmer's past experiences with a specific crop becomes a valuable
resource for making informed predictions in subsequent crop cycles (Lata & Chaudhari, 2019) .
International Journal of Computer Science and Mathematical Theory (IJCSMT) E -ISSN 2545-5699
P-ISSN 2695-1924 Vol 10. No.6 2024 www.iiardjournals.org
IIARD – International Institute of Academic Research and Development
Page 121
Crop yield prediction ventures into the realm of advanced algorithms, forecasting crop production
by harnessing a diverse array of data points. These include temperature, rainfall, pH levels,
pesticide and fertilizer application, and various meteorological variables. The intricacies of crop
yield forecasting hinge on the application of predefined criteria, creating a pathway to anticipate
and enhance the productivity of each crop cycle (Oluwole et al., 2022) . Conventionally, farmers
have predominantly emplo yed conventional farming methods. However, these methods lack
precision, leading to decreased productivity and extensive time consumption. Precision farming,
on the other hand, enhances productivity by identifying the necessary practices for each season.
Elements of precision farming encompass forecasting weather conditions, soil analysis,
recommending suitable crops for cultivation, and determining precise quantities of fertilizers and
pesticides required (Abdulraheem et al., 2022; Durai & Shamili, 2022). Several researchers have
leveraged machine learning methodologies in many parts of the world, including regression trees,
random forests, multivariate regression, association rule mining, and artificial neural networks, to
predict crop yields. Utilizing machine-learning (ML) models, crop -yield is taken as an implicit
function of input variables, encompassing factors such as weather components, and soil conditions,
thereby presenting a potentially complex and nonlinear function. According to (Durai & Shamili,
2022; Mgendi, 2024) in the context of Precision Farming (PF), innovative technologies such as
ML, Data Mining, Data Analytics, and Internet of a Thing (IoT) converge to usher in a new era of
data-driven agricultural efficiency. These innovative tools are not just data collectors; they are
orchestrators of insights, contributing to a reduction in manual labor while concurrently amplifying
productivity. At the heart of this technological symphony, ML techniques take center stage,
autonomously detecting, i dentifying, and predicting outcomes by extracting knowledge , and
relationships from the vast datasets under scrutiny (Domingues et al., 2022). In the agricultural
landscape of Nigeria, as delineated by (Jiya et al., 2023) , crop cultivation emerges as the
predominant pursuit, carrying significant implications for families and the broader economy.
The accurate prediction of crop yields stands as a critical challenge in the agricultural domain,
necessitating advanced AI -driven solutions (Mgendi, 2024) . Farmers confront a plethora of
uncertainties, including limited knowledge about potential harvest excess, unpredictable weather-
conditions, and dynamic seasonal rainfall policies. The intricacies extend to the depletion of soil
nutrition levels, the volatile availability and cost of fertilizers, challenges in pest control, and the
ever-present specter o f post -harvest losses (Oluwole et al., 2022) . These multifaceted factors
collectively contribute to a decline in crop production (Durai & Shamili, 2022). The integration of
AI technologies can provide a sophisticated framework for addressing these challenges
comprehensively. AI algorithms can analyze vast datasets, assimilating information on weather
patterns, soil health, fertilizer dynamics, and pest management. This data-driven approach enables
precise predictions and recommendations, empowering farmers with actionable insights to
optimize crop yield, resource allocation, and risk mitigation. Exploiting AI holds the promise of
enhancing decision-making processes, promoting sustainable practices, and fostering resilience in
the face of diverse challenges (Durai & Shamili, 2022; Mgendi, 2024).
Traditionally, farmers have adhered to conventional farming practices, which lack precision and
result in reduced productivity and time consumption, particularly in the face of unpredictable
climate conditions (Zhou & Ismaeel, 2021 ; Prabavathi & Chelliah, 2022). Exploiting the use of
variable farming resources like soil, fertilizer, and weather data, including crop information, is of
International Journal of Computer Science and Mathematical Theory (IJCSMT) E -ISSN 2545-5699
P-ISSN 2695-1924 Vol 10. No.6 2024 www.iiardjournals.org
IIARD – International Institute of Academic Research and Development
Page 122
utmost importance. Achieving high yields hinges on the efficient deployment of these resources
(Kwaghtyo & Eke, 2023) . In AI -driven agriculture, forecasting weather conditions, scrutinizing
soil properties, suggesting optimal crops for cultivation, and determining precise quantities of
fertilizers and pesticides constitute key facets of precision farming (Zhou & Ismaeel, 2021) .
Exploiting cutting-edge technologies such as IoT, data mini ng, data analytics, and machine -
learning, precise farming systematically gathers data, trains sophisticated systems, and forecasts
outcomes (Mgendi, 2024). This technological integration not only minimizes the need for manual
labor but also enhances overall productivity in the agricultural landscape (Durai & Shamili, 2022;
Mgendi, 2024). Predicting crop yields stands out as a formidable challenge in the domain of smart
agriculture, prompting the proposal and validation of numerous models (Kuradusenge et al., 2023;
Mgendi, 2024). Given that crop production is influenced by various factors like weather, climate,
seed quality, fertilization, and soil composition, addressing this challenge requires the integration
of diverse datasets (Oluwole et al., 2022). Studies by Bali & Garba (2021), Bali et al. (2021) and
Bali (2024) demonstrated that neural network -based methods can effectively learn from
environmental data, self-organize their structures, and deliver strong prediction performance.
Awe and Dias (2022) developed a deep-learning model, specifically a generative Artificial Neural
Networks (ANN), to effectively capture and estimate the complex and non-parametric features of
agricultural output. The study was conducted in Nigeria with data covering forty years (1980 -
2019). They compared the results with the autoregressive integrated moving average model
(ARIMA). The empirical findings demonstrate the superiority of the hybrid ANN model over the
traditional Box -Jenkins ARIMA methodology for forecasting non -stationary time series. The
proposed ANN model significantly improves forecasting accuracy in this context. A fuzzy
inference system (FIS) was developed by (Shuaibu et al., 2021) to forecast rice yield in Jigawa
state, Nigeria, for the period 2021 -2030. This forecasting model incorporated variables such as
“rainfall, land availability, and historical rice production data”. The findings of this study indicate
that in the year 2022, a total land area of 177,950 hectares will be demanded, along with a rainfall
of 873 millimeters, to achieve a rice production of 1,070,000 metric tons. Utilizing an ANN
multilayer perceptron (MLP), Alves et al. (2018) predicted soybean productivity by considering
growth habits, sowing density, and agronomic characteristics. The outcomes reveal a
commendable 98% success rate on the training dataset and a noteworthy 72% accuracy on the
validation dataset, demonstrating the network's capability to estimate soybean productivity based
on agronomic features, growth habits, and pop ulation density. Recognizing the profound impact
of crop production setbacks, this research pivots toward a comparative exploration of diverse
techniques for predicting crop yields.
3. Methodology
3.1.Study Area
This exploration is slated to unfold within the Michika Local Government Area of Adamawa State,
strategically situated in the far northeast, marked by geographic coordinates covering latitude 11°8'
South and longitude 15°13' E. This locality shares a northern boundary with Madagali LGA and
extends further to form an international boundary with the Republic of Cameroon to the northeast.
Borno State bounds it to the West, while Mubi and Hong LGAs define its southern boundary. The
Michika LGA, a mosaic of diversity, encompass es four distinctive development areas: Michika
Metropolitan, Garta, Bazza, and Madzi, sprawled across a land mass measuring 961 km2. The
International Journal of Computer Science and Mathematical Theory (IJCSMT) E -ISSN 2545-5699
P-ISSN 2695-1924 Vol 10. No.6 2024 www.iiardjournals.org
IIARD – International Institute of Academic Research and Development
Page 123
dynamic shade of this region is woven together by an estimated population of 211,124 occupants.
Embarking on a temporal line, this study is projected for a duration of seven months, from May to
November during the Wet Season. This strategic time frame aims to encapsulate and scrutinize the
nuances of environmental dynamics and agricultural practices, offering a comprehensive insight
into the intricacies of the Michika Local Government Area.
3.2.Crop yield parameters
In this area, groundnut, maize, and rice are the predominant crops. The essential weather factors
impacting crop yield include precipitation, air temperature, air moisture, solar radiation, and wind
speed. An AI -driven analysis, guided by , was employed to examine climate and weather
parameters. Additionally, historical crop yield data linked to soil and weather conditions,
encompassing temperature, humidity, rainfall, and soil pH, were employed for the nominated crops
in the region. The equipment used for data collection is shown in Table 1.
Table 1 Showing equipment used for data collection
S/
N
Equipment Use
1 Soil pH Tester (Excellent Corrosion Proof Soil Fertility
Nutrient Meter)
Measures soil pH
2 Digital thermometer Measures temperature
3 Hygrometer (wet and dry bulb) Measures humidity
4 Rain gauge (Programmable Water Timer) Measures rainfall
3.3.Selected Crops
Groundnut (Arachis hypogaea), Maize (Zea mays), and Rice (Oryza sativa) datasets were used.
The accession of crop yield data involved collating information from the agricultural department
and various farmer cooperatives within the study area during the Wet season. The major factors of
the dataset include District, farm size (ha), Fertilizers (Nitrogen, Phosphorus, Potash) (Kg) ,
temperature (oC), rainfall (mm), humidity (RH), yield (kg/ha). May to November was regarded as
the Wet Season. The crop yield in kg/ha was taken as the target variable. The visualization of the
crop yield dataset is shown in Figure 1.
International Journal of Computer Science and Mathematical Theory (IJCSMT) E -ISSN 2545-5699
P-ISSN 2695-1924 Vol 10. No.6 2024 www.iiardjournals.org
IIARD – International Institute of Academic Research and Development
Page 124
Figure 1 presents the visualization of the crop yield dataset.
3.4.DATA PRE-PROCESSING
The data for this exploration passed a series of preprocessing steps aimed at optimizing the
accuracy of crop yield production predictions. MATLAB (2021a) served as the primary
development platform for creating and assessing the proposed model. Initially, a rigorous data
cleaning process was executed, involving transformative activities to exclude any duplicate or
irrelevant data points within the dataset. Following this, feature selection was performed to assess
the correlation between features and the target variable, crop yield. This analysis aids in detecting
and retaining the most pertinent features. Subsequently, feature scaling was employed to
standardize numerical features, employing techniques such as min-max scaling or standardization.
This normalization ensures that all numerical features are on a comparable scale. Concurrently,
categorical variables were converted into numerical representations to enhance compatibility with
machine learning algorithms. In the final stages of preprocessing, the dataset was partitioned into
distinct sets for training, validation, and testing. The training set was employed for model training,
the validation set contributed to hyperparameter tuning, and the testing set served as the benchmark
for estimating the model's performance.
This research explores four categories of machine learning algorithms: ANN, decision trees (DT),
Random Forest (RF), and support vector machines (SVM). The motivation for exploring machine
learning algorithms such as ANN, DT, RF, and SVM stems from their ability to handle complex,
nonlinear connections within data and giv e accurate predictions for diverse agricultural
applications. These algorithms are well -suited to optimize crop yield predictions due to their
inflexibility, scalability, and rigidity in modeling intricate patterns and relations between various
environmental, soil, and management factors. The activity diagram fo r the model is shown in
Figure 2.
International Journal of Computer Science and Mathematical Theory (IJCSMT) E -ISSN 2545-5699
P-ISSN 2695-1924 Vol 10. No.6 2024 www.iiardjournals.org
IIARD – International Institute of Academic Research and Development
Page 125
Figure 2: The activity diagram depicting the flow of tasks and decisions in the crop yield
prediction process.
i) Data Collection: Relevant historical data on crop yields, climate, soil conditions, and other
relevant factors were collected.
ii) Data Preprocessing : The data was prepared and standardized, handling missing values,
scaling, and encoding categorical variables. The data from inputs and target datasets were
standardized between zero and one using (1) to avoid large numeric ranges from the values
of the predictor variables.
minmax
min
nornal
xx
xxx −−=
(1)
where 𝑥 is the respective variable, 𝑥𝑚𝑖𝑛 is the minimum value, 𝑥𝑚𝑎𝑥 is the maximum, and
𝑥𝑛𝑜𝑟𝑚al is the standardized value.
International Journal of Computer Science and Mathematical Theory (IJCSMT) E -ISSN 2545-5699
P-ISSN 2695-1924 Vol 10. No.6 2024 www.iiardjournals.org
IIARD – International Institute of Academic Research and Development
Page 126
iii) Split Data: The data was divided into training and testing datasets.
iv) Model Selection : Each of the machine learning models ANN, DT, RF, and SVM was
designated for training.
v) Model Training: Each model was trained on the training dataset.
vi) Model Evaluation: The performance of each model using appropriate metrics was estimated.
The statistical metrics employed to estimate the performance of artificial intelligence-driven
predictive models are RMSE (Root Mean Squared Error), R -squared, and Mean Absolute
Error (MAE). The motivation for using RMSE, R -squared, and MAE to evaluate models in
crop yield prediction is to give complementary insights into accuracy and reliability. RMSE
measures the average squared error, lower values indicate better model performance, correct
large errors, and assess model accuracy and sensitivity to variability. MAE measures the
average absolute error, lower values indicate better performance, treat all deviations equally,
and balance average accuracy while being robust to outliers. R² indicates the proportion of
variance explained by the model; values closer to 1 indicate better performance, while
negative values suggest the model performs worse than a simple mean predictor. The training
errors were tracked using the RMSE function, which is specified as:
2
n
ni
nettargeti )y(yn1RMSE = −=
(2)
R2, or the relationship between the model ed data value and the observed data value is examined
via the coefficient of determination given by:
2
1
2
12
)(
)(
1
=
−
−
−= n
i
i
n
i
xx
xx
R
(3)
Where
xi = the observed value
x = the modeled value
n = the number of errors,
Σ = summation symbol,
(xi – x)2 = the square of absolute errors.
2R
= the correlation coefficient
x
= the average value
The Mean Absolute Error (MAE) represents the average of the absolute differences between the
actual target values and the predicted values. It is mathematically defined as shown in (4).
−
−=
N
i
ii yyNMAE 1
1
(4)
vii) Hyperparameter Tuning: The model hyperparameters were optimized for improved model
performance.
viii) Prediction: The trained models were used to predict crop yield for new data.
International Journal of Computer Science and Mathematical Theory (IJCSMT) E -ISSN 2545-5699
P-ISSN 2695-1924 Vol 10. No.6 2024 www.iiardjournals.org
IIARD – International Institute of Academic Research and Development
Page 127
ix) Results Visualization: The output was visualized and the predictions were compared with
actual values using charts or plots.
This entire process was implemented in MATLAB using built-in functions; fitcensemble, fitctree,
svm, and trainNetwork functions for each machine learning model, along with built-in evaluation
functions; crossval, cvLoss, and predict.
4. Interpretation of Results and Discussion
The results from the four machine learning models; ANN, DT, RF, and SVM as shown in Table 2
provide valuable insights into the performance of the models in predicting crop yield. The
performance of each model was discussed based on the three metrics: RMSE, MAE, and R².
Table 2 present a clear comparison of how each model performs evaluated using RMSE, MAE,
and R².
Model RMSE MAE R2
DT 5992.0 3841.8 -2.3817
SVM 3315.6 2593.9 -0.035425
RF 3430.6 2733.0 -0.10846
ANN 3136.8 2502.2 0.073255
DT: The high RMSE (5992) value for the DT model indicates that its predictions diverge
significantly from the actual yield values on average. This suggests that the DT could not capture
the complexity of the relationship between the predictors and the target variable well. MAE:
Similarly, the MAE (3841.8) is relatively high, meaning that the average absolute error between
the predicted and actual values is large. This further supports the conclusion that the model is
struggling to predict crop yields accurately. The negative R² (-2.3817) value indicates that the DT
model is performing worse than a simple mean -based model (i.e., always predicting the average
yield). This is a clear sign of poor model performance and suggests that the DT is overfitting or
capturing very little of the variance in the target variable. This s hows that the DT model is
underperforming, likely due to overfitting or the inability to capture the complex patterns in the
data. The negative R² and high RMSE/MAE suggest that this model is not suitable for crop yield
prediction in this case.
RF: with RMSE (3430.6) is slightly worse than SVM, indicating higher error. It is the third best,
slightly worse than SVM with MAE (2733). R² (-0.10846) is negative, meaning the model also
performs worse than the baseline mean predictor. This indicates that RF struggles to generalize
well and has suboptimal performance compared to SVM and ANN.
SVM: This is the s econd best after ANN with RMSE (3315.6), indicating decent predictive
accuracy. Second best after ANN with MAE (2593.9). The R² (0.035425) is close to zero, showing
the model explains very little of the variance and marginally outperforms the baseline. This shows
that SVM performs moderately well but has limited predictive power.
ANN: the l owest RMSE (3136.8), indicates the most accurate predictions. The lowest MAE
(2502.2), is suggesting the smallest average prediction errors. R² (0.073255) is slightly positive,
International Journal of Computer Science and Mathematical Theory (IJCSMT) E -ISSN 2545-5699
P-ISSN 2695-1924 Vol 10. No.6 2024 www.iiardjournals.org
IIARD – International Institute of Academic Research and Development
Page 128
indicating the model explains a small amount of variance and outperforms the baseline predictor.
This indicates that ANN outperforms the other models in all metrics, making it the best -suited
model for this dataset and problem.
5. The Comparison of Models
The bar chart in Figure 3 illustrates the comparison of RMSE values for four different machine -
learning models: DT, SVM, RF, and ANN.
Figure 3 visualizes a comparison of the RSME of the models
DT is the worst -performing model, with high RMSE and MAE values, and a very negative R²,
indicating that the model is poorly fitted to the data and might be overfitting or not capturing the
relevant features of the dataset.
The bar chart in Figure 4 presents a comparison of the MAE values for four different machine-
learning models: DT, SVM, RF, and ANN.
International Journal of Computer Science and Mathematical Theory (IJCSMT) E -ISSN 2545-5699
P-ISSN 2695-1924 Vol 10. No.6 2024 www.iiardjournals.org
IIARD – International Institute of Academic Research and Development
Page 129
Figure 4 visualizes a comparison of the MAE of the models
SVM and RF : Both perform similarly, with SVM slightly outperforming RF in terms of RMSE
and MAE. Both models show negative R² values, indicating they are not suitable for explaining
the variance in crop yields. They are more accurate than DT, but not as effective as ANN. The bar
chart in Figure 5 compares the MAE values across four machine-learning models: DT, SVM, RF,
and ANN.
Figure 5 visualizes a comparison of the R2 of the models
International Journal of Computer Science and Mathematical Theory (IJCSMT) E -ISSN 2545-5699
P-ISSN 2695-1924 Vol 10. No.6 2024 www.iiardjournals.org
IIARD – International Institute of Academic Research and Development
Page 130
Best Performing Model: ANN consistently outperforms the other models in all metrics, followed
by SVM. RF and DT perform poorly, with DT being the worst. Among the models, ANN shows
the best performance in terms of RMSE, MAE, and R². The generally low or negative R² values
indicate that all models, except ANN, struggle to capture the underlying patterns in the data. This
suggests that the available features (Temperature, Rainfall, Humidity, and District) might not fully
capture the complexity of the factors influencing crop yield, or that further feature engineering,
data preprocessing, or model tuning may be required for better predictive accuracy. However,
ANN’s positive R² suggests it is slightly capable of explaining the variance, and the most reliable
model for predicting crop yield.
6. Conclusion
This research evaluates various yield prediction techniques, analyzing their effectiveness to
recommend strategies based on weather parameters and nutrient requirements. The study
developed a predictive model capable of accurately forecasting crop yields using features deduced
from comprehensive datasets. This model offers farmers in the study area actionable insights into
optimal crop varieties, fertilizer application, and prevailing environmental conditions, thereby
enhancing decision -making and reducing potential losses. The findings emphasize the
transformative potential of machine learning (DT, RF, and SVM) in optimizing agricultural
productivity. While the ANN model demonstrated superior performance by minimizing prediction
errors, the overall predictive accuracy was constrained, as indicated by low R² values. Addressing
these limitations presents a n important opportunity for future research, which could incorporate
real-time data from IoT sensors, scale the model to different crops and regions, and develop more
robust, user-friendly tools to promote widespread adoption. Additionally, this study highlights the
importance of optimizing hyperparameters, such as learning rate, number of neurons, and
activation functions, to improve predictive accuracy. Exploring hybrid modeling combining ANN
with other techniques is also recommended, as they have the potential to leverage complementary
strengths of various approaches, thereby enhancing overall model performance.
Acknowledgments
We sincerely appreciate Adamawa State University, Mubi, for providing financial support for this
research through the TETFUND grant. Additionally, we are grateful to the authors of the valuable
resources referenced, the domain experts who contributed their insights, an d the agricultural
department alongside the farmer cooperatives within the study area for their collaboration and
support.
Conflict of Interest
The authors affirm that there are no conflicts of interest associated with this study.
Funding
This research was supported by funding from the Tertiary Education Trust Fund (TETFund) under
the Institution -Based Research (IBR) program, with reference n umber
TEFT/DR&D/UNI/MUBI/RG/2024/VOL.1, facilitated through Adamawa State University, Mubi.
International Journal of Computer Science and Mathematical Theory (IJCSMT) E -ISSN 2545-5699
P-ISSN 2695-1924 Vol 10. No.6 2024 www.iiardjournals.org
IIARD – International Institute of Academic Research and Development
Page 131
References
Abdulraheem, M., Awotunde, J. B., Abidemi, E. A., Idowu, D. O., & Adekola, S. O. (2022).
Weather prediction performance evaluation on selected machine learning algorithms. IAES
International Journal of Artificial Intelligence, 11(4), 1535.
Alves, G. R., Teixeira, I. R., Melo, F. R., Souza, R. T. G., & Silva, A. G. (2018). Estimating
soybean yields with artificial neural networks. Acta Scientiarum. Agronomy, 40.
Awe, O. O., & Dias, R. (2022). Comparative analysis of ARIMA and artificial neural network
techniques for forecasting non -stationary agricultural output time series. AGRIS on-line
Papers in Economics and Informatics, 14(665-2023-114), 3-9.
Bali, B. (2024). Analysis of Emerging Trends in Artificial Intelligence in Education in Nigeria.
Bali, B., & Garba, E. J. (2021). Neuro -fuzzy approach for prediction of neurological disorders: a
systematic review. SN Computer Science, 2(4), 307.
Bali, B., Garba, E. J., & Ahmadu, A. S. (2021). Adaptive Neuro Fuzzy Inference System for
Diagnosis of Stimulant Use Disorders.
Domingues, T., Brandão, T., & Ferreira, J. C. (2022). Machine learning for detection and
prediction of crop diseases and pests: A comprehensive survey. Agriculture, 12(9), 1350.
Durai, S. K. S., & Shamili, M. D. (2022). Smart farming using machine learning and deep learning
techniques. Decision Analytics Journal, 3, 100041.
Jiya, E. A., Illiyasu, U., & Akinyemi, M. (2023). Rice Yield Forecasting: A Comparative Analysis
of Multiple Machine Learning Algorithms. Journal of Information Systems and
Informatics, 5(2), 785-799.
Kuradusenge, M., Hitimana, E., Hanyurwimfura, D., Rukundo, P., Mtonga, K., Mukasine, A., ...
& Uwamahoro, A. (2023). Crop yield prediction using machine learning models: case of
Irish potato and maize. Agriculture, 13(1), 225.
Kwaghtyo, D. K., & Eke, C. I. (2023). Smart farming prediction models for precision agriculture:
a comprehensive survey. Artificial Intelligence Review, 56(6), 5729-5772.
Lata, K., & Chaudhari, B. (2019). Crop yield prediction using data mining techniques and machine
learning models for decision support system. Journal of Emerging Technologies and
Innovative Research (JETIR).
Mgendi, G. (2024). Unlocking the potential of precision agriculture for sustainable
farming. Discover Agriculture, 2(1), 87
Oluwole, O. E., Osaghae, E. O., & Basaky, F. D. (2022). Machine learning Solution for Prediction
of Soil Nutrients for Crop Yield: A Survey. Machine learning, 9(9).
Prabavathi, R., & Chelliah, B. J. (2022). A comprehensive review on machine learning approaches
for yield prediction using essential soil nutrients. Universal Journal of Agricultural
Research, 10(3), 288-303.
International Journal of Computer Science and Mathematical Theory (IJCSMT) E -ISSN 2545-5699
P-ISSN 2695-1924 Vol 10. No.6 2024 www.iiardjournals.org
IIARD – International Institute of Academic Research and Development
Page 132
Shuaibu, A. M., Muhammad, M. N., & Abu -Safyan, Y. (2021). Forecasting Rice Production in
Jigawa State, Nigeria using Fuzzy Inference System. Dutse J. Pure Appl. Sci , 7(4), 203-
213.
Zhou, Q., & Ismaeel, A. (2021). Integration of maximum crop response with machine learning
regression model to timely estimate crop yield. Geo-Spatial Information Science , 24(3),
474-483.
论文原文与出版信息
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE AND MATHEMATICAL THEORY E-ISSN