Precision Farming in Nepal: A Machine Learning Perspective
Abstract
This paper encompasses three different machine learning models that we built to help Nepali farmers in selecting ideal crops for their land, using the right fertilizers, and predicting plant diseases. We tried about five models each for crop recommendation and fertilizer recommendation and a single model for plant disease prediction. We chose “Decision Trees” for both our Crop Recommendation and Fertilizer Recommendation and “Convolutional Neural Networks (CNN)” for Plant Disease Prediction. All models achieved over 95% accuracy. Our GitHub repository houses all the code, making it accessible for future researchers and ML developers working on related tasks. (https://github.com/anamgiri/uunchai).
Keywords
Download Options
Introduction
Most of the foodstuffs in Nepal are still imported from foreign countries, despite the fact that agriculture is the primary occupation for most people. While technology has significantly advanced in other sectors in Nepal, the agricultural sector still relies on traditional farming methods, which are more time-consuming and less productive. Farmers are uneducated about various modern farming practices that could be very beneficial to them. Therefore, we developed machine learning models to assist Nepali farmers in integrating technology into their farming methods. Our focus is solely on the agricultural sector, including both professional farmers and individuals growing crops at home. Through this paper, we aim to demonstrate how our model can identify the right crops to plant under optimal conditions, recommend appropriate fertilizers, and accurately predict plant diseases in a timely manner, which will greatly assist in precision farming for Nepali farmers.
Conclusion
AND FUTURE WORK: In summary, we successfully developed three ML models to assist Nepali farmers in choosing the right crops, selecting appropriate fertilizers, and predicting plant diseases. For the Plant Disease Prediction System, our future work includes using different CNN architectures with early stopping to address potential overfitting. Moving forward, we plan to utilize more complex and pre-trained models to analyze their accuracies. For the Crop and Fertilizer Recommendation Systems, our future efforts will focus on finding datasets with more parameters. We aim to collect larger and updated datasets specific to Nepal, which could revolutionize precision agriculture with technology. Additionally, we plan to integrate these three systems into our website. Currently, the models are only hyperlinked to our site, but in the future, we intend to develop an interactive UIwhere users can input their parameters and receive recommendations or predictions directly.
V. DISCUSSION 5.1 Potential Impact: • Increased Productivity: By optimizing crop selection, fertilizer use, and disease management, these models can significantly boost agricultural yields in Nepal. • Reduced Costs: Precision farming minimizes resource wastage (fertilizers, pesticides) and reduces labor costs by automating certain tasks. • Improved Food Security: Increased productivity can contribute to greater food security for Nepal, potentially reducing reliance on imports. • Sustainability: Optimized resource use can lead to more sustainable agricultural practices, minimizing environmental impact. 5.2 Technological Advancement: • The successful implementation of these models demonstrates the potential of machine learning in modernizing Nepalese agriculture. • This can encourage further research and development in this area, leading to more sophisticated and impactful solutions. 5.3 Farmer Empowerment: By providing farmers with data-driven insights and decision-making tools, these models can empower them to make informed choices and improve their livelihoods.
VI. LIMITATIONS 6.1 Data Limitations: • Data Availability: Relying on datasets from other countries can introduce biases and limit the model'saccuracy in the specific context of Nepal. • Data Quality: The quality of available data significantly impacts model performance. Inaccurate or incomplete data can lead to unreliable predictions. • Data Collection: Collecting high-quality, real-time data from Nepalese farms can be challenging due to limited infrastructure and resources. 6.2 Model Limitations: • Generalization: Models trained on limited datasets may not generalize well to new, unseen situations or variations in environmental conditions. • Interpretability: Some complex models, like deep neural networks, can be difficult to interpret, making it challenging to understand the rationale behind their predictions. • Maintenance: Machine learning models require ongoing maintenance, including retraining with new data and adapting to changing conditions. 6.3 Implementation Challenges: • Technology Access: Ensuring access to technology (smartphones, internet connectivity) for all farmers in Nepal can be a significant hurdle. • Digital Literacy: Farmers may require training and support to effectively use and understand the outputs of these models. • Trust and Adoption: Building trust among farmers in the use of technology and convincing them to adopt new practices can be challenging. 6.4 Addressing Limitations: • Data Collection: Invest in initiatives to collect high-quality, location-specific data on soil, weather, and crop conditions in Nepal. • Model Development: Explore more robust and interpretable models, such as explainable AItechniques. • Technology Access: Improve digital infrastructure in rural areas and provide affordable access to smartphones and internet connectivity. • Farmer Education: Conduct workshops and training programs to educate farmers on the use of these technologies and their benefits. • Continuous Improvement: Regularly monitor model performance, gather feedback from farmers, and continuously refine models based on real-world experience.
