AI-Powered Crop Yield Prediction System Using Multi-Model ML & Real-Time Agricultural Data

Abstract

Agriculture has traditionally relied on farmers’ experiential knowledge to select crops based on seasonal patterns and regional familiarity. Yet, as climate conditions become more unpredictable, soil degradation, and the growing global demand for food, such traditional methods are no longer sufficient. The Nonexistence of timely, data-driven decision-making tools often results in poor crop selection, suboptimal yields, and inefficient resource utilization, particularly in resource-constrained regions. This paper proposes a ML-based CRS designed to analyze essential agricultural parameters such as soil nutrients (nitrogen, phosphorus, potassium), pH level, rainfall, and temperature to identify the most suitable crop for a given environment. The system utilizes supervised learning algorithms including RF, Decision Trees, and SVM, trained on comprehensive historical datasets containing crop yields and environmental profiles. A lightweight web-based interface enables users to input their soil data and receive real-time, region-specific crop recommendations. Experimental evaluation demonstrates high predictive accuracy, with the RF algorithm consistently outperforming others in terms of generalization and reliability across diverse agricultural zones. The proposed solution aids in mitigating crop mismatch risks, promotes sustainable land use, and enhances agricultural productivity through intelligent, data-driven support. The system is scalable and adaptable to make it usable on a larger scale in different areas agro-climatic regions.

Introduction

Agriculture is vital for economic development and livelihoods, especially in countries like India. However, challenges such as unscientific farming practices, climate variability, and poor crop selection reduce productivity and cause financial instability for farmers. Traditional crop selection methods, often based on inherited knowledge, lack the precision needed to adapt to changing environmental conditions.

Proposed Solution:
The study presents a machine learning (ML)-based crop recommendation system that analyzes soil and climatic parameters—like nitrogen, phosphorus, potassium levels, temperature, humidity, pH, and rainfall—to suggest the most suitable crop for cultivation. The system uses algorithms including Random Forest (RF), Decision Trees (DT), Support Vector Machines (SVM), and Naive Bayes, with RF highlighted for its robustness and ability to handle noisy data.

Key Features:

• User-friendly web interface accessible to farmers with minimal technical expertise.

• Offline compatibility for remote areas with limited internet.

• Lightweight design suitable for mobile and low-resource devices.

• Real-time recommendations based on local environmental data.

Literature Review:
Prior research has explored various ML techniques such as Naive Bayes, Decision Trees, SVM, and deep learning for crop prediction and yield forecasting. Ensemble methods like RF consistently show superior accuracy and generalizability. Studies also emphasize the importance of preprocessing, data balancing, and user-centric designs.

Methodology:

• Dataset of 2,200+ agricultural samples with soil nutrients and weather attributes.

• Data cleaning, normalization, and label encoding.

• Model training using stratified sampling, k-fold cross-validation, and hyperparameter tuning.

• Deployment via a Flask-based web application allowing easy user input and instant crop suggestions.

Evaluation and Results:

• RF outperformed SVM, Logistic Regression, and k-NN in accuracy (87.6%) and F1-score (86.4%).

• A CNN-based model achieved even higher accuracy (95.8%) and F1-score (95.1%).

• Precision, recall, F1-score, confusion matrix, and inference time were used to evaluate performance.

• Precision-Recall curves showed strong model capability in handling class imbalance.

• The system enables precise, data-driven crop selection, promoting sustainable farming and better resource utilization.

Significance:
Widespread use of this system can optimize regional cropping strategies, enhance national agricultural output, and support precision agriculture goals by tailoring crop choices to specific environmental conditions.

Conclusion

The proposed Crop Recommendation System presents an effective solution to the challenges faced by farmers in selecting appropriate crops based on environmental and soil conditions. Conventional farming methods usually depend on subjective judgment and limited access to agricultural expertise, which can result in suboptimal crop choices and poor yields. By integrating machine learning into the decision-making process, this system enables precise, data-driven crop recommendations that improve agricultural efficiency and sustainability. The framework utilizes essential parameters such as nitrogen, phosphorus, potassium content, pH, temperature, humidity, and rainfall to predict the most suitable crop for a specific location. Through the application of various supervised learning algorithms including RF,SVM, Logistic Regression, and k-Nearest Neighbors the system were trained and tested on historical agricultural data. Among the models evaluated, the Random Forest algorithm achieved the highest accuracy, exceeding 96%, and demonstrated superior performance across key evaluation metrics such as rigor, remind, and F1-score. This high-performing model was integrated into a user-friendly interface, enabling farmers and agricultural planners to easily input data and receive crop suggestions in real time. Visualizations of model performance and feature importance helped enhance transparency and trust in the system. The deployment of the system ensures that General users with no advanced technical training” can benefit from advanced analytics, promoting equitable access to precision farming insights. Although the system achieves reliable results and practical utility, future enhancements are possible.

References

[1] 2024 Ivan?Malashin et?al. (2024) – Predicting Sustainable Crop Yields: Deep Learning and Explainable AI Tools. Introduces DL and XAI methods to forecast yields, emphasizing model transparency and interpretability Farmonaut®+13MDPI+13MDPI+13. [2] 2023CYP using ML and DL Techniques’ (Proc. Computer Science, 2023) – Applied ML/DL (RF, SVM, LSTM) on five major Indian crops; Random Forest achieved R² ??0.963 ScienceDirect. [3] 2022 Sajid et?al. (2022) – County scale crop yield prediction by integrating crop simulation with ML. Combined crop simulation and ML across the U.S. Corn Belt, achieving ~9% RRMSE Frontiers+1.[3]2021 Fan et?al. (2021) – GNN RNN Approach for Harnessing Geospatial and Temporal Info. Utilized a graph based recurrent model for county level yield prediction across 2000+ U.S. counties Frontiers+4arXiv+4arXiv+4. [4] 2020 van Klompenburg et?al. (2020) –ML forCYP: Systematic review. Comprehensive survey of ML/DL techniques, noting prevalent use of vegetation indices and CNNs ScienceDirect+1. [5] 2019 Khaki &?Wang (2019) –CYP Using DNN. DNNs trained on Syngenta large maize-genotype dataset, outperforming shallow methods arXiv. [6] 2017 Zhou et?al. & Tanabe et?al. (2023 citing works starting in ~2017) – UAV based multispectral imaging combining feature and image based ML/DL for early yield prediction in rice and wheat breeding trials ScienceDirect. [7] 2016 Khaki &?Wang (2019’s historical dataset) – As part of the Syngenta dataset study, their deep learning model evaluated performance through 2016 agricultural seasons arXiv. [8] 2015 Sujatha &?Isakki (2016 described 2015 work) – Yield forecasting using classification techniques (2015). One of the earliest studies using classical classification models.

Copyright

Copyright © 2025 Dr. Girish Kumar D, Miss. Rupa K. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.