Crop classification is a crucial task in modern agriculture, enabling farmers to optimize crop selection based on specific soil and climatic conditions, thereby improving yield and resource eficiency. This study presents a comparative analysis of machine learning techniques for crop classification, utilizing key input features such as soil properties (pH, nitrogen, phosphorus, potassium) and local weather data. Several machine learning models were evaluated for their performance in terms of accuracy, precision, sensitivity, and overall robustness. Among the tested models, the Bagging classifier demonstrated superior performance, achieving an accuracy of 99.77%, making it the most efective approach for the given dataset. The ifndings highlight the significant potential of machine learning in transforming agricultural practices, ofering a data-driven pathway for sustainable crop management. The study also identifies future research opportunities, including the integration of diverse data sources and addressing real-world implementation challenges, to further enhance the applicability and scalability of these techniques.
their distinct attributes, the system achieved an accuracy 2.2. Data Preprocessing
of 95%. Another research [
In the context of region-specific applications, a smart
agricultural system designed for Algerian farmers [
While these studies highlight the potential of ML for
crop classification, identifying the most efective
algorithm remains an open challenge. This study addresses
this gap by performing a comprehensive comparative
analysis of several leading ML algorithms, including
Multi-Layer Perceptron (MLP), Support Vector Machines
(SVM), Decision Trees (DT), Random Forest (RF),
KNearest Neighbors (KNN), Naive Bayes (NB), Stacking,
Bagging, XGBoost, and LightGBM. Leveraging a
publicly available dataset that incorporates critical soil and
climate attributes, we evaluate these models based on
accuracy, adaptability, and computational eficiency.
• Naive Bayes (NB): A probabilistic classifier that
applies Bayes’ theorem to estimate the likelihood
of diferent crop classes based on feature
probabilities [
from Kaggle [
To assess the performance of the employed machine learning models in crop classification, this study utilized a range of evaluation metrics: accuracy, recall, precision, and F1-score. Accuracy measures the overall proportion of correctly classified crop types. Recall focuses on the model’s ability to identify true positives, meaning the proportion of actual positive cases the model correctly predicted. Precision, on the other hand, evaluates the model’s ability to avoid false positives, indicating the proportion of predicted positive cases that were truly positive. Finally, the F1-score provides a harmonic mean between precision and recall, ofering a balanced view of model performance. By considering these metrics together, we gain a comprehensive understanding of the model’s strengths and weaknesses in classifying crop types [41, 42].
Classifiers RF DT NB SVM KNN Raw Data
techniques in enhancing machine learning model performance for crop classification. A systematic evaluation was conducted to determine the impact of various preprocessing steps on commonly used classification models, including Random Forest (RF), Decision Tree (DT), Support 3.1.3. Data Cleaning Vector Machine (SVM), K-Nearest Neighbors (KNN), and Handling missing values and outliers is crucial to ensure Naive Bayes (NB). The preprocessing methods analyzed model reliability. The study compared the performance included dataset splitting strategies, feature selection, of models trained on: data cleaning, and normalization.
tures substantially improved classification accuracy. This underscores the importance of incorporating relevant features to enhance the predictive capabilities of machine learning models for crop classification tasks. 3.1.1. Dataset Splitting Splitting the dataset into training and testing subsets is a fundamental step in machine learning to evaluate model generalization. This study assessed two widely adopted data-splitting ratios: • 50/50 split: Allocates equal portions of the dataset for training and testing. • 80/20 split: Assigns 80% for training and 20% for testing, ensuring a larger training set. removing missing values led to reduced accuracy due to the loss of potentially valuable data. 3.1.4. Data Normalization
mitigating issues caused by varying feature magnitudes. The impact of normalization on model performance is detailed in Table 3. Most models, particularly SVM and KNN, exhibited significant accuracy improvements postnormalization, with SVM achieving a notable 20.54% increase. However, tree-based models like DT and RF showed minimal improvements, reflecting their inherent insensitivity to feature scaling. 3.1.5. Key Insights from Preprocessing Techniques Figure 5 highlight the impact of various preprocessing steps on the accuracy of the model, demonstrating that an 80/20 data split consistently outperformed the 50/50 split by providing a larger training set, particularly for smaller datasets. Normalization significantly improved performance for models sensitive to feature scaling, such as SVM and KNN, but had minimal impact on tree-based models. The inclusion of seven features instead of four led to better classification accuracy, emphasizing the importance of selecting relevant variables. Additionally, median imputation was shown to be the most efective approach for handling missing values, maintaining the integrity and precision of the data set compared to data removal, which led to a performance drop.
This section evaluates the performance of various machine learning algorithms applied to the crop classification task, with an emphasis on analyzing their results before and after hyperparameter tuning. The objective is to identify the optimal model configurations and assess the impact of preprocessing and tuning on model performance.
The dataset was split into 80% training and 20% testing, with model performance measured using standard metrics such as Accuracy, Precision, Recall, and F1-score. 43–50
formance of each algorithm without hyperparameter tuning. Table 4 summarizes the results, showcasing the strengths and weaknesses of the models in their default configurations.
Stacking emerged as the best-performing model with a testing accuracy of 99.54% and perfect scores across all other evaluation metrics. Naive Bayes (NB) followed closely with a testing accuracy of 99.31%, demonstrating strong potential for crop classification. Random Forest (RF) achieved 99.09% testing accuracy, further highlighting the reliability of ensemble methods. Other models, including SVM, KNN, and Decision Trees, performed well but did not surpass these top-performing algorithms.
Figure 2 visually compares the testing and training accuracies, reinforcing these observations. The initial evaluation underscores the inherent strengths of each model and sets a benchmark for further improvement through hyperparameter tuning.
search to optimize model parameters systematically. Table 5 lists the selected hyperparameters for each model, which were fine-tuned to enhance performance.
Following hyperparameter tuning, we re-evaluated the performance of each model on the test set. Table 6 summarizes the obtained results. The impact of hyperparameter tuning is evident across all models, with significant improvements observed in testing accuracy, precision, recall, and F1-score. Figure 3: Testing Accuracy of Models (Hyperparameter Tun
The results are impressive. Across all evaluation met- ing). rics, we observed significant gains in performance for each model (Figures 3 & 4). Notably, the Bagging ensemble classifier emerged as the champion, achieving a maintained a high accuracy of 99.31% while achieving remarkable testing accuracy of 99.77%. This indicates perfect scores (100%) for Precision, Recall, and F1-score. that the Bagging ensemble, by combining multiple deci- This suggests their exceptional ability to correctly idension trees with optimized hyperparameters, efectively tify both positive and negative instances (specific crop learned complex patterns within the crop data and deliv- types) within the data. ered outstanding classification accuracy. The positive impact of hyperparameter tuning is evi
Following Bagging closely were Random Forest (RF) dent across all models (Figure 5). For instance, SVM saw and Stacking, both reaching an accuracy of 99.54%. This a significant accuracy improvement of 1.14%, reaching highlights the efectiveness of ensemble methods and 99.09%. Similarly, the Decision Tree (DT) benefitted from combining multiple models for crop classification tasks. tuning, with its accuracy increasing by 0.23% to 98.41%.
It’s also worth noting the consistent performance of These improvements showcase the power of hyperpaNaive Bayes (NB) and XGBoost (XGB). These models rameter optimization in unlocking the full potential of
The research presented here compared various machine learning algorithms for crop classification. Notably, Bagging classifiers emerged as the frontrunner, achieving an impressive testing accuracy of 99.77%. This superior performance, coupled with high precision and recall, translates to the potential for significant advancements in several key areas: Agricultural Productivity, Resource Optimization, and Sustainable Food Systems.
Building upon this foundation, future research should explore integrating additional data sources, such as satellite imagery or real-time sensor data from fields. Additionally, investigating real-world implementation challenges to ensure accessibility and adoption by farmers will be crucial in realizing the transformative potential of this technology.
ChatGPT, Grammarly in order to: Grammar and spelling check, Paraphrase and reword. After using this tool/service, the authors reviewed and edited the content as needed and take full responsibility for the publication’s content.