Current IssueSpecial IssueEditorial BoardOnline First ArticlesArchive
Current IssueSpecial IssueEditorial BoardOnline First ArticlesArchive

Full Research Article

Data-driven Agriculture: Enhancing Crop Prediction with NPK Sensor Integration in Semi-arid Karnataka, India

V
V.M. Aparanji1
P
Praveen Kumar Y.G.2,*
A
Ashwini S. Shivannavar2
S
Sujata N. Patil3
M
Mohammed Kaif1
1Department of Electronics and Communication Engineering, Siddaganga Institute of Technology, Tumakuru 572 103, Karnataka, India.
2Department of Electronics and Communication Engineering, Sri Siddhartha Institute of Technology, SSAHE, Maraluru, Tumakuru 572 105, Karnataka, India.
3Department of Electronics and Communication Engineering, KLE Technological University, BVB College Campus, Hubli-580 031, Karnataka, India.

ABSTRACT

Background: Agriculture supports over 2.5 billion livelihoods worldwide, yet soil fertility management on Indian smallholder farms remains largely empirical. Tumakuru district’s red soil compound this problem with low organic matter and high spatial variability in N, P and K. Conventional lab testing, costs ₹350-450 and takes 3-7 days, putting it out of reach for most smallholders. Further, existing studies rarely validate sensors against lab references, report seasonal accuracy to actionable fertilization guidance. To address these issues, a low-cost edge-computing system integrated with soil sensors and machine learning is proposed.

Methods: An NPK sensor and DHT22 sensor were interfaced with a Raspberry Pi 3B+ and deployed at 18 plots in Tumakuru district across kharif (June-October 2023) and rabi (November 2022-March 2023) seasons. Sensor readings were validated against laboratory reference methods (Kjeldahl for N; Olsen for P; ammonium acetate for K; n = 20 composite samples). A field-collected dataset of 3,600 instances was classified by Decision Tree and K-Nearest Neighbour (KNN) classifiers under 10-fold stratified cross-validation. A novel Confidence-Weighted Majority Voting (CWMV) ensemble was evaluated by McNemar’s test.

Result: Sensor laboratory correlations were strong (N: r = 0.91; P: r = 0.87; K: r = 0.89; RMSE≤9.3%). Cross-validated accuracy was 96.7±0.4% (Decision Tree) and 96.5±0.5% (KNN). The CWMV ensemble achieved 97.3% test-set accuracy with statistically significant improvement. Seasonal accuracy varied by 0.4 percentage points. Hardware cost was ₹4,200 (≈US$50; ₹0.19 per prediction; 99.95% cheaper than laboratory testing) and energy consumption was 0.101 Wh per 1,000 inference cycles.

INTRODUCTION

Agriculture supports more than 2.5 billion livelihoods worldwide, yet soil fertility management on smallholder farms in India remains largely empirical (Van Klompenburg et al., 2020; Medar et al., 2019, Venugopal et al., 2021; Anonymous, 2023). Red laterite soils dominating Tumakuru district, Karnataka -characterised by low organic matter (0.4-0.8%) and high spatial variability in N, P and K-demand accurate crop-soil matching for sustainable yields (Velayutham et al., 1999 and Anonymous, 2017). Conventional soil testing (Kjeldahl for N; Olsen for P; ammonium acetate for K) costs ₹350-450 per sample and takes 3-7 days, placing it beyond the reach of most smallholders (Kapoor, 2025).
       
Precision agriculture bridges this gap by deploying in-situ NPK sensors and machine learning (ML) classifiers on low-power microcontrollers (edge computing), as demonstrated in several recent studies (Islam et al., 2023; Senapaty et al., 2023; Manju et al., 2024; Senapaty et al., 2024; Warpe et al., 2024; Saha et al., 2025). Islam et al., (2023) showed that IoT-based soil nutrient monitoring combined with ML achieves reliable crop recommendations without laboratory support. Senapaty et al. (2023, 2024) demonstrated that IoT-enabled soil nutrient analysis with Support Vector Machine classifiers achieves high accuracy for crop recommendation on resource-constrained hardware. However, three critical gaps persist across the literature: sensor readings are rarely validated against laboratory reference methods (Van Klompenburg et al., 2020 and Mahendra et al., 2020); seasonal prediction accuracy across kharif and rabi cycles is unreported (Islam et al., 2023; Manju et al., 2024 and Saha et al., 2025); and agronomic interpretation linking ML predictions to fertilisation actions is largely absent (Anonymous, 2017; Babu et al., 2024).
       
The specific objectives of this work is to: (i) validate the RS-NPKHNWSYF-N01 sensor against laboratory methods for Tumakuru red laterite soils; (ii) assemble a field-collected, season-stratified dataset of 3,600 soil profiles across four crops; (iii) develop and statistically validate a novel CWMV ensemble; (iv) quantify seasonal accuracy, hardware cost and energy consumption; and (v) interpret predictions against ICAR Karnataka fertilisation guidelines (Anonymous, 2017).

MATERIALS AND METHODS

This work is carried out in the department of electronics and communication engineering, Siddaganga Institute of Technology and Sri Siddhartha Institute of Technology (SSAHE), Tumakuru, Karnataka, India and took around 15 months to carry out this research work.
 
Study area, data collection and sensor validation
 
Field data were collected from 18 agricultural plots in Tumakuru district, Karnataka, India (13.34°N, 77.10°E; 820 m a.s.l.) across the kharif season (June-October 2023; n = 1,980 records from 12 plots) and rabi season (November 2022-March 2023; n = 1,620 records from 10 plots); 8 plots were active in both seasons. The NPK sensor (RS-NPKHNWSYF-N01; Renke, China) measured N, P and K (mg kg-1) via electrical resistance and RS485 communication (9600 baud), converted to TTL for Raspberry Pi GPIO interfacing (Islam et al., 2023; Manju et al., 2024). A capacitive soil moisture sensor, DHT22 (humidity/temperature) and MCP3008 ADC completed the sensor suite (Senapaty et al., 2023). System architecture and workflow are shown in Fig 1 and Fig 2.

Fig 1: System architecture.



Fig 2: Data collection and prediction workflow.


 
Data pre-processing and dataset
 
Raw RS485 frames were parsed, validated and standardised (z-score normalisation) by a Python (v3.9) pipeline (Pedregosa et al., 2011). 14 sensor-fault records were discarded and < 0.3% missing values imputed by median. The cleaned dataset (n = 3,600; four balanced classes, n = 900 per crop) was partitioned by stratified random sampling into training (75%; n = 2,700) and held-out test (25%; n = 900) sets (Elbasi et al., 2023; Gupta et al., 2023). Pre-processed soil data sample for different crops is illustrated in Table 1. The feature correlation heatmap is shown in Fig 3.

Fig 3: Pearson correlation heatmap of six input features (N, P, K, temperature, humidity, moisture; n = 3,600).



Table 1: Pre-processed soil data sample.


 
Machine learning classifiers and CWMV ensemble
 
Two classifiers were selected for interpretability and computational suitability for edge deployment (Senapaty et al., 2023; Senapaty et al., 2024). The decision tree (Pedregosa et al., 2011) recursively minimises Gini impurity (Equation 1); KNN assigns the majority class among k = 5 nearest neighbours (Equation 2), with k selected by 5-fold cross-validation (k∈ {3, 5, 7, 9, 11}) (Pedregosa et al., 2011; Elbasi et al., 2023). Both were evaluated by 10-fold stratified cross-validation (mean ± SD) (Elbasi et al., 2023; Gupta et al., 2023; Ramaiah et al., 2023). The novel CWMV ensemble resolves classifier disagreements using KNN posterior class probability P_KNN(y|x) and threshold π = 0.70 (Equations 3), selected by 5-fold CV scanning π∈ {0.55-0.80}. The CWMV was evaluated on the held-out test set (Manju et al., 2024); improvement over alternatives was assessed by McNemar’s test (continuity-corrected, α = 0.05).

Gini(t) = 1 − Σᵢ p(i|t)²   ... (1)

ŷ = argmax_y Σᵢ₌₁ᵏ I(yᵢ = y)   ... (2)

ŷ_CWMV = ŷ_DT if ŷ_DT = ŷ_KNN;  ŷ_KNN if ŷ_DT ≠ ŷ_KNN and P_KNN ≥ τ;  ŷ_DT otherwise   ... (3)
 
Confusion matrices are shown in Table 2. Sensor accuracy was validated against accredited laboratory methods (Department of Soil Science, University of Agricultural Sciences, Dharwad) using 20 composite samples (0-20 cm depth; three sub-samples blended per composite) collected simultaneously with sensor readings (Saha et al., 2025). Reference methods were: Kjeldahl for available N; Olsen method (0.5 M NaHCO3) for available P; and ammonium acetate extraction (1 M, pH 7.0) for available K (Anonymous, 2017). Pearson r and RMSE were calculated per nutrient, with pre-defined acceptability thresholds of r ≥0.80 and RMSE ≤10% of the laboratory mean illustrated in Table 3.

Table 2: Confusion matrices for (a) Decision Tree , (b) KNN and (c) CWMV ensemble on the test set (n = 900).



Table 3: NPK sensor validation against laboratory methods for Tumakuru red laterite soils (n = 20).


 
System architecture, cost and energy profiling
 
The Raspberry Pi 3B+ (ARM Cortex-A53 @ 1.4 GHz; 1 GB RAM; Raspberry Pi OS Lite) performs all inference locally (<200 ms per cycle) (Islam et al., 2023; Senapaty et al., 2023), displaying predictions on a 16×2 I2C LCD and a Flask web dashboard over local Wi-Fi as illustrated in Fig 4. Power consumption was measured by a USB power meter (Ruideng UM25C) over 1,000 consecutive cycles at steady state. Energy per 1,000 cycles: E1000 = V × I_mean × (t × 1000/3600) = 5.00 V × 0.390 A × 0.0519 = 0.101 Wh. Statistical analyses used one-way ANOVA with Tukey HSD (crop-class separability), McNemar’s test (classifier comparisons) (Manju, 2024) and Clopper-Pearson 95% confidence intervals (seasonal accuracy), all via Python scipy v1.11 (Pedregosa et al., 2011).

Fig 4: Flask web dashboard showing real-time sensor readings (N, P, K, moisture, humidity, temperature) and the CWMV crop prediction for an example field measurement.

RESULTS AND DISCUSSION

Sensor validation results as illustrated in Table 3 confirmed strong correlations against laboratory reference methods for all three nutrients (N: r = 0.91; P: r = 0.87; K: r = 0.89; all p<0.001), with RMSE ranging from 6.1% (N) to 9.3% (K) of the laboratory mean-within the 10% acceptability threshold. The first published validation of this instrument class against laboratory methods for Indian semi-arid soils as stated by Van Klompenburg et al. (2020). The lower P correlation (r = 0.87) reflects the known sensitivity of available phosphorus to instantaneous soil pH and moisture (Anonymous, 2017), which can cause divergence from the time-averaged Olsen extraction.
       
Crop-class separability (Table 4) was confirmed by one-way ANOVA with Tukey HSD: N (F3,3596 = 5,412; p<0.001), P (F3,3596  = 2,871; p<0.001) and K (F3,3596 = 3,104; p < 0.001) (Pedregosa et al., 2011). The mango-coconut K pair showed the smallest effect (p = 0.003), explaining most classifier errors between these classes (Kalimuthu et al., 2020 and Elbasi et al., 2023). Feature correlation analysis (Fig 3) confirmed N-K (r = 0.85) and N-P (r = 0.71) as the strongest pairs, consistent with balanced NPK fertilisation practices in the district Velayutham et al., 1999 and Anonymous, 2017).

Table 4: Crop-specific NPK threshold profiles (mean ± SD, n = 900/crop).


       
Classifier performance (Table 5): 10-fold cross-validation yielded Decision Tree accuracy of 96.7±0.4% and KNN accuracy of 96.5±0.5% (Pedregosa et al., 2011). These results are consistent with prior ML crop prediction studies reporting accuracies in the 85-97% range (Elbasi et al., 2023; Gupta et al., 2023; Ramaiah et al., 2023; Musanase et al., 2023; Prity et al., 2024). On the held-out test set, the CWMV ensemble achieved 97.3%, reducing total misclassifications from 30 (Decision Tree) and 27 (KNN) to 24. McNemar’s test confirmed statistically significant improvement over both individual classifiers (p = 0.02 and p = 0.03 respectively). The 27 vs 24 error improvement over naïve voting was not significant (p = 0.09), but all three corrected errors were mango-coconut confusions-the agronomically highest-consequence pair (Section 4.1). Mean inference latency was 187 ms (n = 100 cycles) (Islam et al., 2023 and Senapaty et al., 2023). Flask web dashboard showing real-time sensor readings is shown in Fig 4.

Table 5: Classification performance: 10-fold CV (mean ± SD, training set) for individual classifiers; test set (n = 900) for ensemble methods.


       
Seasonal robustness (Table 6): kharif CWMV accuracy was 97.1% (95% CI: 95.6-98.3%) and rabi 96.7% (95% CI: 94.9-98.0%); the 0.4 percentage-point difference was not statistically significant (McNemar’s test, p = 0.31), confirming year-round reliability. This seasonal consistency has not been previously reported for comparable sensor-based and ML based crop prediction systems (Islam et al., 2023; Manju et al., 2024; Saha et al., 2025; Metagar et al., 2024; Hassan et al., 2026). Hardware and energy results (Table 7): total hardware cost was ₹4,200 (≈US$50), yielding ≈ ₹0.19 per prediction-99.95% less than laboratory testing (Kapoor et al., 2025). Energy consumption was 0.101 Wh per 1,000 inference cycles, enabling approximately 81 hours on a 10,000 mAh battery or 24-hour autonomy with a 5 W solar panel under Karnataka’s average solar irradiance (Saha et al., 2025).

Table 6: Seasonal CWMV accuracy and per-crop per-season instance distribution (Clopper-pearson 95% CI).



Table 7: Bill of materials (BOM) and per-prediction cost vs laboratory testing (INR; 1 USD » 84 INR; supplier invoices, April 2024).


 
Sensor validity and agronomic implications
 
The validated NPK threshold profiles (Table 4) can be directly mapped to ICAR fertilisation recommendations (Anonymous, 2017). A ‘banana’ prediction (N > 100, K > 55 mg kg-1) signals a nutrient-rich soil requiring balanced P supplementation rather than blanket NPK application (ICAR recommendation: 200–220 kg N ha-1 (Anonymous, 2017)). A ‘maize’ prediction (N < 50, K < 30 mg kg-1) indicates nitrogen deficiency requiring urea at 120-150 kg N ha-1 (Anonymous, 2017). ‘Mango’ or ‘coconut’ predictions indicate moderate NPK status for perennial orchard establishment (Anonymous, 2017); management diverges sharply between these species -coconut requires 500 g N palm-1 yr-1 and 7-10 years to full production, while mango reaches production in 3-5 years at 50-100 kg N ha-1 (Anonymous, 2017).
       
The mango-coconut confusion carries the greatest agronomic risk of all classifier errors (Anonymous, 2017), as a misclassification could direct a farmer to a substantially more capital-intensive, longer-gestation planting decision. The CWMV ensemble’s targeted reduction of these confusions from 11 to 7 on the test set is therefore disproportionately valuable relative to its overall accuracy increment. Future work should implement a cost-sensitive classifier that formally penalises this confusion pair (Dey et al., 2024).
 
Accuracy compared with prior work
 
The CWMV accuracy of 97.3% and individual classifier accuracies of 96.-96.7% are directly comparable with Manju et al., (2024), who achieved 97.3% KNN on a four-crop sensor-collected Indian dataset with the same 75:25 split -providing cross-state empirical validation. Higher accuracies reported by Dey et al., (2024) (XGBoost, 99.09% on Kaggle benchmark) and Rani et al. (2023) (Gradient Boosting, 99.27%) were obtained on curated benchmark datasets, which consistently overestimate field performance as noted by Saha et al., (2025). The accuracy of 97.3% is also consistent with findings of Elbasi et al., (2023) (Bayes Net, 99.59% on benchmark) and surpasses the 90.4% and 92.1% reported for KNN and SVM by Ramaiah et al., (2023) on similar Indian multi-crop datasets. The narrow cross-validation SDs (0.4-0.5%) confirm the robustness of these estimates (Pedregosa et al., 2011). The sensor validation results address a foundational gap identified by Van Klompenburg et al. (2020) in their 50-study systematic review: in-situ sensor measurements are rarely validated against laboratory reference methods before use in classifier training.
 
Practical deployment and limitations
 
At ₹0.19 per prediction-99.95% less than laboratory testing (Kapoor et al., 2025) -the system is economically accessible to smallholder farmers. The edge-computing architecture, requiring no cloud connectivity, outperforms cloud-dependent systems reviewed by Islam et al. (2023) and Senapaty et al. (2023, 2024) in connectivity-limited settings. Energy consumption of 0.101 Wh per 1,000 cycles confirms solar-powered autonomous deployment is feasible (Saha et al., 2025). The study has several limitations: it covers four crops in one district; sensor calibration was conducted on 20 samples from one soil order (Velayutham et al., 1999); and the CWMV improvement over naïve voting was not statistically significant (p = 0.09). Multi-region validation-including temperate crops relevant to broader agricultural audiences (Anonymous, 2023 and Musanase et al., 2023) is needed before broader generalisation. Future work will expand the sensor suite to include soil pH, Ca and Mg (Dey et al., 2024); broaden the crop portfolio (Elbasi et al., 2023; Gupta et al., 2023); implement SHAP explainability (Rani et al., 2023); and develop a cost-sensitive classifier penalising agronomically high-consequence misclassification pairs (Babu et al., 2024).

CONCLUSION

The study demonstrated the feasibility of a sensor-validated, edge-deployed ML system for crop prediction in semi-arid Karnataka. The RS-NPKHNWSYF-N01 NPK sensor was validated against laboratory reference methods (r = 0.87-0.91; RMSE≤9.3%)-the first published calibration for this instrument class in Indian semi-arid conditions. A 3,600-instance, two-season field dataset supported 10-fold cross-validated classifiers (Decision Tree: 96.7±0.4%; KNN: 96.5±0.5%), with the CWMV ensemble achieving 97.3% test accuracy and statistically significant improvement over individual classifiers (McNemar: p≤ 0.03). Seasonal accuracy varied by 0.4 percentage points (p = 0.31), confirming year-round reliability-a finding not previously reported for comparable sensor-based systems. Predictions were mapped to ICAR fertilisation recommendations, with mango–coconut confusion identified as the highest-consequence misclassification pair. The system costs ₹4,200 (≈US$50), delivers predictions at ₹0.19 each and consumes 0.101 Wh per 1,000 inference cycles-confirming feasibility for solar-powered permanent deployment. Subject to multi-region validation, this system offers a promising, agronomically grounded framework for precision agriculture decision support for smallholder farmers.

ACKNOWLEDGEMENT

The authors thank the farming communities of Tumakuru district and the Department of Soil Science, University of Agricultural Sciences, Dharwad, for laboratory analyses.
 
Disclaimers
 
The views and conclusions expressed in this article are solely those of the authors and do not necessarily represent the views of their affiliated institutions.

CONFLICT OF INTEREST

The authors declare that there are no conflicts of interest regarding the publication of this article. No funding or sponsorship influenced the design of the study, data collection, analysis, decision to publish, or preparation of the manuscript.

REFERENCES


  1. Anonymous, Food and Agriculture Organization of the United Nations. (2023). The state of food and agriculture 2023: Revealing the true cost of food to transform agrifood systems. https://doi.org/10.4060/cc7724en.

  2. Anonymous, Indian Council of Agricultural Research. (2017). Crop production guide for Karnataka. ICAR.

  3. Babu, G.R., Gokuldhev, M. and Brahmanandam, P.S. (2024). Integrating IoT for soil monitoring and hybrid machine learning in predicting tomato crop disease in a typical South India station. Sensors. 24(19): 61-77. https://doi.org/10.3390/ s24196177.

  4. Dey, B., Ferdous, J. and Ahmed, R. (2024). Machine learning based recommendation of agricultural and horticultural crop farming in India under the regime of NPK, soil pH and three climatic variables. Heliyon. 10(3): e25112. https:// doi.org/10.1016/j.heliyon.2024.e25112.

  5. Elbasi, E., Zaki, C., Topcu, A.E., Abdelbaki, W., Zreikat, A.I., Cina, E., Shdefat, A. and Saker, L. (2023). Crop prediction model using machine learning algorithms. Applied Sciences. 13(16): https://doi.org/10.3390/app13169288.

  6. Gupta, T., Maggu, S. and Kapoor, B. (2023). Crop prediction using machine learning. Iconic Research and Engineering Journals. 6(9): 279-284.

  7. Hassan N. Bharathi, Hiran, D. and Tripathy, A.K. (2026). Spatio- temporal assessment of vegetation and NDVI time series forecasting model using remote sensing and machine learning for Karnataka, India. Agricultural Science Digest. doi: 10.18805/ag.D-6498.

  8. Islam, M.R., Oliullah, K., Kabir, M.M., Alom, M. and Mridha, M.F. (2023). Machine learning enabled IoT system for soil nutrients monitoring and crop recommendation. Journal of Agriculture and Food Research. 14: 100880. https://doi.org/10.1016/ j.jafr.2023.100880.

  9. Kalimuthu, M., Vaishnavi, P. and Kishore, M. (2020). Crop prediction using machine learning. In 2020 Third International Conference on Smart Systems and Inventive Technology (ICSSIT). 926-932. IEEE. https://doi.org/10.1109/ICSSIT48 917.2020.9214190.

  10. Kapoor, S., Pal, B.D., Singhal, A. and Anantha, K.H. (2025). Technology intensification and farmers’ welfare: A case study from Karnataka, a semi-arid state of India. The Indian Economic Journal. 73(3): 445-468. https://doi.org/10.1177/00194 662241253871.

  11. Manju, G., Syam Kishor, K.S. and Binson, V.A. (2024). An IoT- enabled real-time crop prediction system using soil fertility analysis. Eng. 5(4): 2496-2510. https://doi.org/10.3390/ eng5040130.

  12. Mahendra, N., Vishwakarma, D., Nischitha, K., Ashwini and Manjuraju, M.R. (2020). Crop prediction using machine learning approaches. International Journal of Engineering Research and Technology. 9(8): 23-26. https://doi.org/10.17577/ IJERTV9IS080029.

  13. Medar, R., V.S. and Shweta, S. (2019). Crop Yield Prediction using Machine Learning Techniques. In 2019 IEEE 5th International Conference for Convergence in Technology (I2CT). pp. 1-5. IEEE. https://doi.org/10.1109/I2CT45611.2019.903 3611.

  14. Metagar, S.M., Walikar, G.A. (2024). Machine learning models for plant disease prediction and detection: A review. Agricultural Science Digest. 44(4): 591-602. doi: 10.18805/ag.D-5893.

  15. Musanase, C., Vodacek, A., Hanyurwimfura, D., Uwitonze, A. and Kabandana, I. (2023). Data-driven analysis and machine learning-based crop and fertilizer recommendation system for revolutionizing farming practices. Agriculture. 13(11): 2141. https://doi.org/10.3390/agriculture13112141.

  16. Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M. et al. (2011). Scikit-learn: Machine learning in Python. Journal of Machine Learning Research. 12: 2825-2830.

  17. Prity, F.S., Hasan, M.M., Saif, S.H., Hossain, M.M., Bhuiyan, S.H., Islam, M.A. and Lavlu, M.T.H. (2024). Enhancing agricultural productivity: A machine learning approach to crop recommendations. Human-Centric Intelligent Systems. 4: 497-510. https://doi.org/10.1007/s44230-024-00081-3.

  18. Ramaiah, D., Bharathi, N. and Devi, R.M. (2023). Machine learning- based crop prediction. International Journal of Innovative Research in Technology. 9(9): 540-544.

  19. Rani, S., Mishra, A.K., Kataria, A., Mallik, S. and Qin, H. (2023). Machine learning-based optimal crop selection system in smart agriculture. Scientific Reports. 13: 15997. https:/ /doi.org/10.1038/s41598-023-42356-y.

  20. Saha, G., Shahrin, F., Khan, F.H., Meshkat, M.M. and Azad, A.A.M. (2025). Smart IoT-driven precision agriculture: Land mapping, crop prediction and irrigation system. PLOS ONE. 20(3): e0319268. https://doi.org/10.1371/journal.pone. 0319268.

  21. Senapaty, M.K., Ray, A. and Padhy, N. (2023). IoT-enabled soil nutrient analysis and crop recommendation model for precision agriculture. Computers. 12(3): 61. https://doi.org/10.33 90/computers12030061.

  22. Senapaty, M.K., Ray, A. and Padhy, N. (2024). A decision support system for crop recommendation using machine learning classification algorithms. Agriculture. 14(8): 1256. https:/ /doi.org/10.3390/agriculture14081256.

  23. Van Klompenburg, T., Kassahun, A. and Catal, C. (2020). Crop yield prediction using machine learning: A systematic literature review. Computers and Electronics in Agriculture. 177: 105709. https://doi.org/10.1016/j.compag.2020.105709

  24. Velayutham, M., Mandal, D. K., Mandal, C., Sehgal, J., Sharma, R. C. and Bhattacharyya, T. (1999). Soil and Land Resources Assessment (NBSS Publication No. 54). National Bureau of Soil Survey and Land Use Planning.

  25. Venugopal, A., Aparna, S., Mani, J., Mathew, R. and Williams, V. (2021). Crop yield prediction using machine learning algorithms. International Journal of Engineering Research and Technology (IJERT). 9(13): 87-91. https://doi.org/10.175 77/IJERTCONV9IS13019.

  26. Warpe, S.T. and Pippal, R.K.S. (2024). A Lagrange interpolation application for automating fertilizer distribution in agriculture using wireless sensor networks. Agricultural Science Digest. 44(4): 618-624. doi: 10.18805/ag.D-5586.
Published In
Agricultural Science Digest
Article Metrics
Metrics Icon
24
Views
0
Citations
Reviewed By
Share Article
Download Article
In this Article
    Contact us
    Follow us

    Full Research Article

    Data-driven Agriculture: Enhancing Crop Prediction with NPK Sensor Integration in Semi-arid Karnataka, India

    V
    V.M. Aparanji1
    P
    Praveen Kumar Y.G.2,*
    A
    Ashwini S. Shivannavar2
    S
    Sujata N. Patil3
    M
    Mohammed Kaif1
    1Department of Electronics and Communication Engineering, Siddaganga Institute of Technology, Tumakuru 572 103, Karnataka, India.
    2Department of Electronics and Communication Engineering, Sri Siddhartha Institute of Technology, SSAHE, Maraluru, Tumakuru 572 105, Karnataka, India.
    3Department of Electronics and Communication Engineering, KLE Technological University, BVB College Campus, Hubli-580 031, Karnataka, India.

    ABSTRACT

    Background: Agriculture supports over 2.5 billion livelihoods worldwide, yet soil fertility management on Indian smallholder farms remains largely empirical. Tumakuru district’s red soil compound this problem with low organic matter and high spatial variability in N, P and K. Conventional lab testing, costs ₹350-450 and takes 3-7 days, putting it out of reach for most smallholders. Further, existing studies rarely validate sensors against lab references, report seasonal accuracy to actionable fertilization guidance. To address these issues, a low-cost edge-computing system integrated with soil sensors and machine learning is proposed.

    Methods: An NPK sensor and DHT22 sensor were interfaced with a Raspberry Pi 3B+ and deployed at 18 plots in Tumakuru district across kharif (June-October 2023) and rabi (November 2022-March 2023) seasons. Sensor readings were validated against laboratory reference methods (Kjeldahl for N; Olsen for P; ammonium acetate for K; n = 20 composite samples). A field-collected dataset of 3,600 instances was classified by Decision Tree and K-Nearest Neighbour (KNN) classifiers under 10-fold stratified cross-validation. A novel Confidence-Weighted Majority Voting (CWMV) ensemble was evaluated by McNemar’s test.

    Result: Sensor laboratory correlations were strong (N: r = 0.91; P: r = 0.87; K: r = 0.89; RMSE≤9.3%). Cross-validated accuracy was 96.7±0.4% (Decision Tree) and 96.5±0.5% (KNN). The CWMV ensemble achieved 97.3% test-set accuracy with statistically significant improvement. Seasonal accuracy varied by 0.4 percentage points. Hardware cost was ₹4,200 (≈US$50; ₹0.19 per prediction; 99.95% cheaper than laboratory testing) and energy consumption was 0.101 Wh per 1,000 inference cycles.

    INTRODUCTION

    Agriculture supports more than 2.5 billion livelihoods worldwide, yet soil fertility management on smallholder farms in India remains largely empirical (Van Klompenburg et al., 2020; Medar et al., 2019, Venugopal et al., 2021; Anonymous, 2023). Red laterite soils dominating Tumakuru district, Karnataka -characterised by low organic matter (0.4-0.8%) and high spatial variability in N, P and K-demand accurate crop-soil matching for sustainable yields (Velayutham et al., 1999 and Anonymous, 2017). Conventional soil testing (Kjeldahl for N; Olsen for P; ammonium acetate for K) costs ₹350-450 per sample and takes 3-7 days, placing it beyond the reach of most smallholders (Kapoor, 2025).
           
    Precision agriculture bridges this gap by deploying in-situ NPK sensors and machine learning (ML) classifiers on low-power microcontrollers (edge computing), as demonstrated in several recent studies (Islam et al., 2023; Senapaty et al., 2023; Manju et al., 2024; Senapaty et al., 2024; Warpe et al., 2024; Saha et al., 2025). Islam et al., (2023) showed that IoT-based soil nutrient monitoring combined with ML achieves reliable crop recommendations without laboratory support. Senapaty et al. (2023, 2024) demonstrated that IoT-enabled soil nutrient analysis with Support Vector Machine classifiers achieves high accuracy for crop recommendation on resource-constrained hardware. However, three critical gaps persist across the literature: sensor readings are rarely validated against laboratory reference methods (Van Klompenburg et al., 2020 and Mahendra et al., 2020); seasonal prediction accuracy across kharif and rabi cycles is unreported (Islam et al., 2023; Manju et al., 2024 and Saha et al., 2025); and agronomic interpretation linking ML predictions to fertilisation actions is largely absent (Anonymous, 2017; Babu et al., 2024).
           
    The specific objectives of this work is to: (i) validate the RS-NPKHNWSYF-N01 sensor against laboratory methods for Tumakuru red laterite soils; (ii) assemble a field-collected, season-stratified dataset of 3,600 soil profiles across four crops; (iii) develop and statistically validate a novel CWMV ensemble; (iv) quantify seasonal accuracy, hardware cost and energy consumption; and (v) interpret predictions against ICAR Karnataka fertilisation guidelines (Anonymous, 2017).

    MATERIALS AND METHODS

    This work is carried out in the department of electronics and communication engineering, Siddaganga Institute of Technology and Sri Siddhartha Institute of Technology (SSAHE), Tumakuru, Karnataka, India and took around 15 months to carry out this research work.
     
    Study area, data collection and sensor validation
     
    Field data were collected from 18 agricultural plots in Tumakuru district, Karnataka, India (13.34°N, 77.10°E; 820 m a.s.l.) across the kharif season (June-October 2023; n = 1,980 records from 12 plots) and rabi season (November 2022-March 2023; n = 1,620 records from 10 plots); 8 plots were active in both seasons. The NPK sensor (RS-NPKHNWSYF-N01; Renke, China) measured N, P and K (mg kg-1) via electrical resistance and RS485 communication (9600 baud), converted to TTL for Raspberry Pi GPIO interfacing (Islam et al., 2023; Manju et al., 2024). A capacitive soil moisture sensor, DHT22 (humidity/temperature) and MCP3008 ADC completed the sensor suite (Senapaty et al., 2023). System architecture and workflow are shown in Fig 1 and Fig 2.

    Fig 1: System architecture.



    Fig 2: Data collection and prediction workflow.


     
    Data pre-processing and dataset
     
    Raw RS485 frames were parsed, validated and standardised (z-score normalisation) by a Python (v3.9) pipeline (Pedregosa et al., 2011). 14 sensor-fault records were discarded and < 0.3% missing values imputed by median. The cleaned dataset (n = 3,600; four balanced classes, n = 900 per crop) was partitioned by stratified random sampling into training (75%; n = 2,700) and held-out test (25%; n = 900) sets (Elbasi et al., 2023; Gupta et al., 2023). Pre-processed soil data sample for different crops is illustrated in Table 1. The feature correlation heatmap is shown in Fig 3.

    Fig 3: Pearson correlation heatmap of six input features (N, P, K, temperature, humidity, moisture; n = 3,600).



    Table 1: Pre-processed soil data sample.


     
    Machine learning classifiers and CWMV ensemble
     
    Two classifiers were selected for interpretability and computational suitability for edge deployment (Senapaty et al., 2023; Senapaty et al., 2024). The decision tree (Pedregosa et al., 2011) recursively minimises Gini impurity (Equation 1); KNN assigns the majority class among k = 5 nearest neighbours (Equation 2), with k selected by 5-fold cross-validation (k∈ {3, 5, 7, 9, 11}) (Pedregosa et al., 2011; Elbasi et al., 2023). Both were evaluated by 10-fold stratified cross-validation (mean ± SD) (Elbasi et al., 2023; Gupta et al., 2023; Ramaiah et al., 2023). The novel CWMV ensemble resolves classifier disagreements using KNN posterior class probability P_KNN(y|x) and threshold π = 0.70 (Equations 3), selected by 5-fold CV scanning π∈ {0.55-0.80}. The CWMV was evaluated on the held-out test set (Manju et al., 2024); improvement over alternatives was assessed by McNemar’s test (continuity-corrected, α = 0.05).

    Gini(t) = 1 − Σᵢ p(i|t)²   ... (1)

    ŷ = argmax_y Σᵢ₌₁ᵏ I(yᵢ = y)   ... (2)

    ŷ_CWMV = ŷ_DT if ŷ_DT = ŷ_KNN;  ŷ_KNN if ŷ_DT ≠ ŷ_KNN and P_KNN ≥ τ;  ŷ_DT otherwise   ... (3)
     
    Confusion matrices are shown in Table 2. Sensor accuracy was validated against accredited laboratory methods (Department of Soil Science, University of Agricultural Sciences, Dharwad) using 20 composite samples (0-20 cm depth; three sub-samples blended per composite) collected simultaneously with sensor readings (Saha et al., 2025). Reference methods were: Kjeldahl for available N; Olsen method (0.5 M NaHCO3) for available P; and ammonium acetate extraction (1 M, pH 7.0) for available K (Anonymous, 2017). Pearson r and RMSE were calculated per nutrient, with pre-defined acceptability thresholds of r ≥0.80 and RMSE ≤10% of the laboratory mean illustrated in Table 3.

    Table 2: Confusion matrices for (a) Decision Tree , (b) KNN and (c) CWMV ensemble on the test set (n = 900).



    Table 3: NPK sensor validation against laboratory methods for Tumakuru red laterite soils (n = 20).


     
    System architecture, cost and energy profiling
     
    The Raspberry Pi 3B+ (ARM Cortex-A53 @ 1.4 GHz; 1 GB RAM; Raspberry Pi OS Lite) performs all inference locally (<200 ms per cycle) (Islam et al., 2023; Senapaty et al., 2023), displaying predictions on a 16×2 I2C LCD and a Flask web dashboard over local Wi-Fi as illustrated in Fig 4. Power consumption was measured by a USB power meter (Ruideng UM25C) over 1,000 consecutive cycles at steady state. Energy per 1,000 cycles: E1000 = V × I_mean × (t × 1000/3600) = 5.00 V × 0.390 A × 0.0519 = 0.101 Wh. Statistical analyses used one-way ANOVA with Tukey HSD (crop-class separability), McNemar’s test (classifier comparisons) (Manju, 2024) and Clopper-Pearson 95% confidence intervals (seasonal accuracy), all via Python scipy v1.11 (Pedregosa et al., 2011).

    Fig 4: Flask web dashboard showing real-time sensor readings (N, P, K, moisture, humidity, temperature) and the CWMV crop prediction for an example field measurement.

    RESULTS AND DISCUSSION

    Sensor validation results as illustrated in Table 3 confirmed strong correlations against laboratory reference methods for all three nutrients (N: r = 0.91; P: r = 0.87; K: r = 0.89; all p<0.001), with RMSE ranging from 6.1% (N) to 9.3% (K) of the laboratory mean-within the 10% acceptability threshold. The first published validation of this instrument class against laboratory methods for Indian semi-arid soils as stated by Van Klompenburg et al. (2020). The lower P correlation (r = 0.87) reflects the known sensitivity of available phosphorus to instantaneous soil pH and moisture (Anonymous, 2017), which can cause divergence from the time-averaged Olsen extraction.
           
    Crop-class separability (Table 4) was confirmed by one-way ANOVA with Tukey HSD: N (F3,3596 = 5,412; p<0.001), P (F3,3596  = 2,871; p<0.001) and K (F3,3596 = 3,104; p < 0.001) (Pedregosa et al., 2011). The mango-coconut K pair showed the smallest effect (p = 0.003), explaining most classifier errors between these classes (Kalimuthu et al., 2020 and Elbasi et al., 2023). Feature correlation analysis (Fig 3) confirmed N-K (r = 0.85) and N-P (r = 0.71) as the strongest pairs, consistent with balanced NPK fertilisation practices in the district Velayutham et al., 1999 and Anonymous, 2017).

    Table 4: Crop-specific NPK threshold profiles (mean ± SD, n = 900/crop).


           
    Classifier performance (Table 5): 10-fold cross-validation yielded Decision Tree accuracy of 96.7±0.4% and KNN accuracy of 96.5±0.5% (Pedregosa et al., 2011). These results are consistent with prior ML crop prediction studies reporting accuracies in the 85-97% range (Elbasi et al., 2023; Gupta et al., 2023; Ramaiah et al., 2023; Musanase et al., 2023; Prity et al., 2024). On the held-out test set, the CWMV ensemble achieved 97.3%, reducing total misclassifications from 30 (Decision Tree) and 27 (KNN) to 24. McNemar’s test confirmed statistically significant improvement over both individual classifiers (p = 0.02 and p = 0.03 respectively). The 27 vs 24 error improvement over naïve voting was not significant (p = 0.09), but all three corrected errors were mango-coconut confusions-the agronomically highest-consequence pair (Section 4.1). Mean inference latency was 187 ms (n = 100 cycles) (Islam et al., 2023 and Senapaty et al., 2023). Flask web dashboard showing real-time sensor readings is shown in Fig 4.

    Table 5: Classification performance: 10-fold CV (mean ± SD, training set) for individual classifiers; test set (n = 900) for ensemble methods.


           
    Seasonal robustness (Table 6): kharif CWMV accuracy was 97.1% (95% CI: 95.6-98.3%) and rabi 96.7% (95% CI: 94.9-98.0%); the 0.4 percentage-point difference was not statistically significant (McNemar’s test, p = 0.31), confirming year-round reliability. This seasonal consistency has not been previously reported for comparable sensor-based and ML based crop prediction systems (Islam et al., 2023; Manju et al., 2024; Saha et al., 2025; Metagar et al., 2024; Hassan et al., 2026). Hardware and energy results (Table 7): total hardware cost was ₹4,200 (≈US$50), yielding ≈ ₹0.19 per prediction-99.95% less than laboratory testing (Kapoor et al., 2025). Energy consumption was 0.101 Wh per 1,000 inference cycles, enabling approximately 81 hours on a 10,000 mAh battery or 24-hour autonomy with a 5 W solar panel under Karnataka’s average solar irradiance (Saha et al., 2025).

    Table 6: Seasonal CWMV accuracy and per-crop per-season instance distribution (Clopper-pearson 95% CI).



    Table 7: Bill of materials (BOM) and per-prediction cost vs laboratory testing (INR; 1 USD » 84 INR; supplier invoices, April 2024).


     
    Sensor validity and agronomic implications
     
    The validated NPK threshold profiles (Table 4) can be directly mapped to ICAR fertilisation recommendations (Anonymous, 2017). A ‘banana’ prediction (N > 100, K > 55 mg kg-1) signals a nutrient-rich soil requiring balanced P supplementation rather than blanket NPK application (ICAR recommendation: 200–220 kg N ha-1 (Anonymous, 2017)). A ‘maize’ prediction (N < 50, K < 30 mg kg-1) indicates nitrogen deficiency requiring urea at 120-150 kg N ha-1 (Anonymous, 2017). ‘Mango’ or ‘coconut’ predictions indicate moderate NPK status for perennial orchard establishment (Anonymous, 2017); management diverges sharply between these species -coconut requires 500 g N palm-1 yr-1 and 7-10 years to full production, while mango reaches production in 3-5 years at 50-100 kg N ha-1 (Anonymous, 2017).
           
    The mango-coconut confusion carries the greatest agronomic risk of all classifier errors (Anonymous, 2017), as a misclassification could direct a farmer to a substantially more capital-intensive, longer-gestation planting decision. The CWMV ensemble’s targeted reduction of these confusions from 11 to 7 on the test set is therefore disproportionately valuable relative to its overall accuracy increment. Future work should implement a cost-sensitive classifier that formally penalises this confusion pair (Dey et al., 2024).
     
    Accuracy compared with prior work
     
    The CWMV accuracy of 97.3% and individual classifier accuracies of 96.-96.7% are directly comparable with Manju et al., (2024), who achieved 97.3% KNN on a four-crop sensor-collected Indian dataset with the same 75:25 split -providing cross-state empirical validation. Higher accuracies reported by Dey et al., (2024) (XGBoost, 99.09% on Kaggle benchmark) and Rani et al. (2023) (Gradient Boosting, 99.27%) were obtained on curated benchmark datasets, which consistently overestimate field performance as noted by Saha et al., (2025). The accuracy of 97.3% is also consistent with findings of Elbasi et al., (2023) (Bayes Net, 99.59% on benchmark) and surpasses the 90.4% and 92.1% reported for KNN and SVM by Ramaiah et al., (2023) on similar Indian multi-crop datasets. The narrow cross-validation SDs (0.4-0.5%) confirm the robustness of these estimates (Pedregosa et al., 2011). The sensor validation results address a foundational gap identified by Van Klompenburg et al. (2020) in their 50-study systematic review: in-situ sensor measurements are rarely validated against laboratory reference methods before use in classifier training.
     
    Practical deployment and limitations
     
    At ₹0.19 per prediction-99.95% less than laboratory testing (Kapoor et al., 2025) -the system is economically accessible to smallholder farmers. The edge-computing architecture, requiring no cloud connectivity, outperforms cloud-dependent systems reviewed by Islam et al. (2023) and Senapaty et al. (2023, 2024) in connectivity-limited settings. Energy consumption of 0.101 Wh per 1,000 cycles confirms solar-powered autonomous deployment is feasible (Saha et al., 2025). The study has several limitations: it covers four crops in one district; sensor calibration was conducted on 20 samples from one soil order (Velayutham et al., 1999); and the CWMV improvement over naïve voting was not statistically significant (p = 0.09). Multi-region validation-including temperate crops relevant to broader agricultural audiences (Anonymous, 2023 and Musanase et al., 2023) is needed before broader generalisation. Future work will expand the sensor suite to include soil pH, Ca and Mg (Dey et al., 2024); broaden the crop portfolio (Elbasi et al., 2023; Gupta et al., 2023); implement SHAP explainability (Rani et al., 2023); and develop a cost-sensitive classifier penalising agronomically high-consequence misclassification pairs (Babu et al., 2024).

    CONCLUSION

    The study demonstrated the feasibility of a sensor-validated, edge-deployed ML system for crop prediction in semi-arid Karnataka. The RS-NPKHNWSYF-N01 NPK sensor was validated against laboratory reference methods (r = 0.87-0.91; RMSE≤9.3%)-the first published calibration for this instrument class in Indian semi-arid conditions. A 3,600-instance, two-season field dataset supported 10-fold cross-validated classifiers (Decision Tree: 96.7±0.4%; KNN: 96.5±0.5%), with the CWMV ensemble achieving 97.3% test accuracy and statistically significant improvement over individual classifiers (McNemar: p≤ 0.03). Seasonal accuracy varied by 0.4 percentage points (p = 0.31), confirming year-round reliability-a finding not previously reported for comparable sensor-based systems. Predictions were mapped to ICAR fertilisation recommendations, with mango–coconut confusion identified as the highest-consequence misclassification pair. The system costs ₹4,200 (≈US$50), delivers predictions at ₹0.19 each and consumes 0.101 Wh per 1,000 inference cycles-confirming feasibility for solar-powered permanent deployment. Subject to multi-region validation, this system offers a promising, agronomically grounded framework for precision agriculture decision support for smallholder farmers.

    ACKNOWLEDGEMENT

    The authors thank the farming communities of Tumakuru district and the Department of Soil Science, University of Agricultural Sciences, Dharwad, for laboratory analyses.
     
    Disclaimers
     
    The views and conclusions expressed in this article are solely those of the authors and do not necessarily represent the views of their affiliated institutions.

    CONFLICT OF INTEREST

    The authors declare that there are no conflicts of interest regarding the publication of this article. No funding or sponsorship influenced the design of the study, data collection, analysis, decision to publish, or preparation of the manuscript.

    REFERENCES


    1. Anonymous, Food and Agriculture Organization of the United Nations. (2023). The state of food and agriculture 2023: Revealing the true cost of food to transform agrifood systems. https://doi.org/10.4060/cc7724en.

    2. Anonymous, Indian Council of Agricultural Research. (2017). Crop production guide for Karnataka. ICAR.

    3. Babu, G.R., Gokuldhev, M. and Brahmanandam, P.S. (2024). Integrating IoT for soil monitoring and hybrid machine learning in predicting tomato crop disease in a typical South India station. Sensors. 24(19): 61-77. https://doi.org/10.3390/ s24196177.

    4. Dey, B., Ferdous, J. and Ahmed, R. (2024). Machine learning based recommendation of agricultural and horticultural crop farming in India under the regime of NPK, soil pH and three climatic variables. Heliyon. 10(3): e25112. https:// doi.org/10.1016/j.heliyon.2024.e25112.

    5. Elbasi, E., Zaki, C., Topcu, A.E., Abdelbaki, W., Zreikat, A.I., Cina, E., Shdefat, A. and Saker, L. (2023). Crop prediction model using machine learning algorithms. Applied Sciences. 13(16): https://doi.org/10.3390/app13169288.

    6. Gupta, T., Maggu, S. and Kapoor, B. (2023). Crop prediction using machine learning. Iconic Research and Engineering Journals. 6(9): 279-284.

    7. Hassan N. Bharathi, Hiran, D. and Tripathy, A.K. (2026). Spatio- temporal assessment of vegetation and NDVI time series forecasting model using remote sensing and machine learning for Karnataka, India. Agricultural Science Digest. doi: 10.18805/ag.D-6498.

    8. Islam, M.R., Oliullah, K., Kabir, M.M., Alom, M. and Mridha, M.F. (2023). Machine learning enabled IoT system for soil nutrients monitoring and crop recommendation. Journal of Agriculture and Food Research. 14: 100880. https://doi.org/10.1016/ j.jafr.2023.100880.

    9. Kalimuthu, M., Vaishnavi, P. and Kishore, M. (2020). Crop prediction using machine learning. In 2020 Third International Conference on Smart Systems and Inventive Technology (ICSSIT). 926-932. IEEE. https://doi.org/10.1109/ICSSIT48 917.2020.9214190.

    10. Kapoor, S., Pal, B.D., Singhal, A. and Anantha, K.H. (2025). Technology intensification and farmers’ welfare: A case study from Karnataka, a semi-arid state of India. The Indian Economic Journal. 73(3): 445-468. https://doi.org/10.1177/00194 662241253871.

    11. Manju, G., Syam Kishor, K.S. and Binson, V.A. (2024). An IoT- enabled real-time crop prediction system using soil fertility analysis. Eng. 5(4): 2496-2510. https://doi.org/10.3390/ eng5040130.

    12. Mahendra, N., Vishwakarma, D., Nischitha, K., Ashwini and Manjuraju, M.R. (2020). Crop prediction using machine learning approaches. International Journal of Engineering Research and Technology. 9(8): 23-26. https://doi.org/10.17577/ IJERTV9IS080029.

    13. Medar, R., V.S. and Shweta, S. (2019). Crop Yield Prediction using Machine Learning Techniques. In 2019 IEEE 5th International Conference for Convergence in Technology (I2CT). pp. 1-5. IEEE. https://doi.org/10.1109/I2CT45611.2019.903 3611.

    14. Metagar, S.M., Walikar, G.A. (2024). Machine learning models for plant disease prediction and detection: A review. Agricultural Science Digest. 44(4): 591-602. doi: 10.18805/ag.D-5893.

    15. Musanase, C., Vodacek, A., Hanyurwimfura, D., Uwitonze, A. and Kabandana, I. (2023). Data-driven analysis and machine learning-based crop and fertilizer recommendation system for revolutionizing farming practices. Agriculture. 13(11): 2141. https://doi.org/10.3390/agriculture13112141.

    16. Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M. et al. (2011). Scikit-learn: Machine learning in Python. Journal of Machine Learning Research. 12: 2825-2830.

    17. Prity, F.S., Hasan, M.M., Saif, S.H., Hossain, M.M., Bhuiyan, S.H., Islam, M.A. and Lavlu, M.T.H. (2024). Enhancing agricultural productivity: A machine learning approach to crop recommendations. Human-Centric Intelligent Systems. 4: 497-510. https://doi.org/10.1007/s44230-024-00081-3.

    18. Ramaiah, D., Bharathi, N. and Devi, R.M. (2023). Machine learning- based crop prediction. International Journal of Innovative Research in Technology. 9(9): 540-544.

    19. Rani, S., Mishra, A.K., Kataria, A., Mallik, S. and Qin, H. (2023). Machine learning-based optimal crop selection system in smart agriculture. Scientific Reports. 13: 15997. https:/ /doi.org/10.1038/s41598-023-42356-y.

    20. Saha, G., Shahrin, F., Khan, F.H., Meshkat, M.M. and Azad, A.A.M. (2025). Smart IoT-driven precision agriculture: Land mapping, crop prediction and irrigation system. PLOS ONE. 20(3): e0319268. https://doi.org/10.1371/journal.pone. 0319268.

    21. Senapaty, M.K., Ray, A. and Padhy, N. (2023). IoT-enabled soil nutrient analysis and crop recommendation model for precision agriculture. Computers. 12(3): 61. https://doi.org/10.33 90/computers12030061.

    22. Senapaty, M.K., Ray, A. and Padhy, N. (2024). A decision support system for crop recommendation using machine learning classification algorithms. Agriculture. 14(8): 1256. https:/ /doi.org/10.3390/agriculture14081256.

    23. Van Klompenburg, T., Kassahun, A. and Catal, C. (2020). Crop yield prediction using machine learning: A systematic literature review. Computers and Electronics in Agriculture. 177: 105709. https://doi.org/10.1016/j.compag.2020.105709

    24. Velayutham, M., Mandal, D. K., Mandal, C., Sehgal, J., Sharma, R. C. and Bhattacharyya, T. (1999). Soil and Land Resources Assessment (NBSS Publication No. 54). National Bureau of Soil Survey and Land Use Planning.

    25. Venugopal, A., Aparna, S., Mani, J., Mathew, R. and Williams, V. (2021). Crop yield prediction using machine learning algorithms. International Journal of Engineering Research and Technology (IJERT). 9(13): 87-91. https://doi.org/10.175 77/IJERTCONV9IS13019.

    26. Warpe, S.T. and Pippal, R.K.S. (2024). A Lagrange interpolation application for automating fertilizer distribution in agriculture using wireless sensor networks. Agricultural Science Digest. 44(4): 618-624. doi: 10.18805/ag.D-5586.
    In this Article
      Contact us
      Follow us
      Published In
      Agricultural Science Digest

      Editorial Board

      View all (0)