A Multiple-Input Neural Network Model for Predicting Cotton Production Quantity: A Case Study
Abstract
:1. Introduction
2. Related Work
3. Data
4. Proposed Multiple-Input Neural Network Prediction Model
- Dense layer: Dense layers [20] constitute the traditional and the most popular choice for composing a hidden layer in a multilayer neural network. A dense layer is composed of neurons which are connected with all neurons in the previous layer. The operations performed by each neuron can be summarized by
- LSTM layer: Long Short-Term Memory (LSTM) layers [21] constitute a special type of Recurrent Neural Networks layers which are characterized by their ability to learn long-term dependencies.Each LSTM unit in the layer is composed of a memory cell and three gates: input, output, and forget. At time t, the input gate and a second gate modulate the amount of information that are stored into the memory state . The forget gate modulates the past information which must be vanished or must be kept on the memory cell at the previous time . Finally, the hidden state constitutes the output of the memory cell and it is calculated using memory state and the output gate which modulates the information used for the output of the memory cell. In summary, the following equations describe the operations performed by an LSTM unit:
- GRU layer: Gated Recurrent Units (GRU) were originally proposed by Cho et al. [22] and were inspired by the LSTM units, but with simpler implementation and calculations. Its main difference is that a GRU unit only has two gates (update and reset) which modulate the flow of information, without the utilization of memory gates, since it exposes the full hidden content without any control.The update gate controls the level the unit updates its content, by taking into consideration a linear sum between the previous state and the input . The reset gate is computed in a similar manner with the update gate . Finally, the activation of a GRU unit constitutes the linear combination between the previous and the candidate activation , which is computed similarly to the traditional recurrent unit. The operations performed by an GRU unit are briefly described by
- MNN utilizes three dense layers of 16, 24, and 10 neurons in Hidden Layer -1-, Hidden Layer -2-, and Hidden Layer -3-, respectively, and a dense layer of 10 neurons after the concatenate layer.
- MNN utilizes three LSTM layers of 30, 50, and 20 units in Hidden Layer -1-, Hidden Layer -2-, and Hidden Layer -3-, respectively, and a dense layer of 12 neurons after the concatenate layer.
- MNN utilizes three GRU layers of 16, 24, and 10 units in Hidden Layer -1-, Hidden Layer -2-, and Hidden Layer -3-, respectively, and a dense layer of 12 neurons after the concatenate layer.
5. State-of-the-Art Machine Learning Models
- DTR is a decision tree dedicated for regression problems, which constructs a model tree based on splitting criteria. More analytically, this algorithm develops a tree with decision nodes and leaf nodes, in which the leaves predict the output continuous value utilizing the linear regression algorithm.
- GP constitutes a collection of random variables depending on time or space, in which every collection of those variables has a multivariate normal distribution. The prediction value of this machine learning algorithm is a one-dimensional Gaussian distribution, and it is calculated by the similarity between the training instances and the unseen instances.
- kNN is a popular machine learning algorithm, which utilizes various distance mathematic formulas to compute feature similarity between each new instance and a predefined number k of instances in the training data. For regression problems, the output value is defined by the average value of its k nearest neighbors.
- LASSO is a linear model trained with prior as a regularizer. This algorithm performs both regularization and variable selection in order to enhance the prediction accuracy. Due to its simplicity and efficiency, it has been successfully extended and applied to a wide variety of statistical models.
- LR probably constitutes the most commonly used algorithm for developing an efficient regression model. This prediction algorithm aims to determine the relationship between one or more explanatory (independent) variables and the dependent variable based on the linear mathematical model.
- SVR is a classical machine learning algorithm which is utilized for predicting continuous values. Its main objective is to fit the error within a specified threshold, in contrast to traditional regression algorithms like LR, which focuses on minimizing the error.
6. Numerical Experiments
7. Discussion and Conclusions
Author Contributions
Funding
Conflicts of Interest
References
- Kalivas, D.; Kollias, V. Effects of soil, climate and cultivation techniques on cotton yield in Central Greece, using different statistical methods. Agronomie 2001, 21, 73–90. [Google Scholar] [CrossRef] [Green Version]
- Chlingaryan, A.; Sukkarieh, S.; Whelan, B. Machine learning approaches for crop yield prediction and nitrogen status estimation in precision agriculture: A review. Comput. Electron. Agric. 2018, 151, 61–69. [Google Scholar] [CrossRef]
- Hayat, A.; Amin, M.; Afzal, S. Statistical investigation to explore the impact of soil and other characteristics on cotton yield. Commun. Soil Sci. Plant Anal. 2020, 51, 1434–1442. [Google Scholar] [CrossRef]
- Sawan, Z.M. Climatic variables: Evaporation, sunshine, relative humidity, soil and air temperature and its adverse effects on cotton production. Inf. Process. Agric. 2018, 5, 134–148. [Google Scholar] [CrossRef]
- Dai, J.; Li, W.; Zhang, D.; Tang, W.; Li, Z.; Lu, H.; Kong, X.; Luo, Z.; Xu, S.; Xin, C. Competitive yield and economic benefits of cotton achieved through a combination of extensive pruning and a reduced nitrogen rate at high plant density. Field Crop. Res. 2017, 209, 65–72. [Google Scholar] [CrossRef]
- Ahmad, S.; Hasanuzzaman, M. Cotton Production and Uses: Agronomy, Crop Protection, and Postharvest Technologies; Springer Nature: London, UK, 2020. [Google Scholar]
- Jabran, K.; Chauhan, B.S. Cotton Production; John Wiley & Sons: Hoboken, NJ, USA, 2019. [Google Scholar]
- Khan, M.A.; Wahid, A.; Ahmad, M.; Tahir, M.T.; Ahmed, M.; Ahmad, S.; Hasanuzzaman, M. World Cotton Production and Consumption: An Overview. In Cotton Production and Uses; Springer: Berlin, Germany, 2020; pp. 1–7. [Google Scholar]
- Ullah, K.; Khan, N.; Usman, Z.; Ullah, R.; Saleem, F.Y.; Shah, S.A.I.; Salman, M. Impact of temperature on yield and related traits in cotton genotypes. J. Integr. Agric. 2016, 15, 678–683. [Google Scholar] [CrossRef] [Green Version]
- Zonta, J.H.; Brandão, Z.N.; Sofiatti, V.; Bezerra, J.R.C.; Medeiros, J.d.C. Irrigation and nitrogen effects on seed cotton yield, water productivity and yield response factor in semi-arid environmernt. Aust. J. Crop. Sci. 2016, 10, 118–126. [Google Scholar]
- Papageorgiou, E.I.; Markinos, A.T.; Gemtos, T.A. Fuzzy cognitive map based approach for predicting yield in cotton crop production as a basis for decision support system in precision agriculture application. Appl. Soft Comput. 2011, 11, 3643–3657. [Google Scholar] [CrossRef]
- Papageorgiou, E.I.; Markinos, A.; Gemptos, T. Application of fuzzy cognitive maps for cotton yield management in precision farming. Expert Syst. Appl. 2009, 36, 12399–12413. [Google Scholar] [CrossRef]
- Jamuna, K.; Karpagavalli, S.; Vijaya, M.; Revathi, P.; Gokilavani, S.; Madhiya, E. Classification of seed cotton yield based on the growth stages of cotton crop using machine learning techniques. In Proceedings of the 2010 International Conference on Advances in Computer Engineering, Bangalore, India, 21–22 June 2010; pp. 312–315. [Google Scholar]
- Haghverdi, A.; Washington-Allen, R.A.; Leib, B.G. Prediction of cotton lint yield from phenology of crop indices using artificial neural networks. Comput. Electron. Agric. 2018, 152, 186–197. [Google Scholar] [CrossRef]
- Nguyen, L.H.; Zhu, J.; Lin, Z.; Du, H.; Yang, Z.; Guo, W.; Jin, F. Spatial-temporal Multi-Task Learning for Within-field Cotton Yield Prediction. In Pacific-Asia Conference on Knowledge Discovery and Data Mining; Springer: Cham, Switzerland, 2019; pp. 343–354. [Google Scholar]
- Brownlee, J. Machine Learning Mastery with Python: Understand Your Data, Create Accurate Models, and Work Projects End-to-End; Machine Learning Mastery: Melbourne, Australia, 2016. [Google Scholar]
- Chandrashekar, G.; Sahin, F. A survey on feature selection methods. Comput. Electr. Eng. 2014, 40, 16–28. [Google Scholar] [CrossRef]
- Khalid, S.; Khalil, T.; Nasreen, S. A survey of feature selection and feature extraction techniques in machine learning. In Proceedings of the 2014 Science and Information Conference, Warsaw, Poland, 7–10 September 2014; pp. 372–378. [Google Scholar]
- Benesty, J.; Chen, J.; Huang, Y.; Cohen, I. Pearson correlation coefficient. In Noise Reduction in Speech Processing; Springer: Berlin/Heidelberg, Germany, 2009; pp. 1–4. [Google Scholar]
- Demuth, H.B.; Beale, M.H.; De Jess, O.; Hagan, M.T. Neural Network Design; Martin Hagan: Stillwater, OK, USA, 2014. [Google Scholar]
- Hochreiter, S.; Schmidhuber, J. Long short-term memory. Neural Comput. 1997, 9, 1735–1780. [Google Scholar] [CrossRef] [PubMed]
- Cho, K.; Van Merriënboer, B.; Bahdanau, D.; Bengio, Y. On the properties of neural machine translation: Encoder-decoder approaches. arXiv 2014, arXiv:1409.1259. [Google Scholar]
- Loh, W.Y. Classification and regression tree methods. In Wiley StatsRef: Statistics Reference Online; Wiley: Hoboken, NJ, USA, 2014. [Google Scholar]
- Paul, W.; Baschnagel, J. Stochastic Processes; Springer: Berlin, Germany, 2013; Volume 1. [Google Scholar]
- Aha, D. Lazy Learning; Springer Science & Business Media: Berlin/Heidelberg, Germany, 2013. [Google Scholar]
- Hastie, T.; Tibshirani, R.; Wainwright, M. Statistical Learning with Sparsity: The Lasso and Generalizations; CRC Press: Boca Raton, FL, USA, 2015. [Google Scholar]
- Seber, G.A.; Lee, A.J. Linear Regression Analysis; John Wiley & Sons: Hoboken, NJ, USA, 2012; Volume 329. [Google Scholar]
- Deng, N.; Tian, Y.; Zhang, C. Support Vector Machines: Optimization Based Theory, Algorithms, and Extensions; Chapman and Hall/CRC: Boca Raton, FL, USA, 2012. [Google Scholar]
- Abadi, M.; Barham, P.; Chen, J.; Chen, Z.; Davis, A.; Dean, J.; Devin, M.; Ghemawat, S.; Irving, G.; Isard, M. Tensorflow: A system for large-scale machine learning. In Proceedings of the 12th USENIX Symposium on Operating Systems Design and Implementation (OSDI’ 16), Savannah, GA, USA, 2–4 November 2016; pp. 265–283. [Google Scholar]
- Pedregosa, F.; Varoquaux, G.; Gramfort, A.; Michel, V.; Thirion, B.; Grisel, O.; Blondel, M.; Prettenhofer, P.; Weiss, R.; Dubourg, V. Scikit-learn: Machine learning in Python. J. Mach. Learn. Res. 2011, 12, 2825–2830. [Google Scholar]
- Dey, R.; Salemt, F.M. Gate-variants of gated recurrent unit (GRU) neural networks. In Proceedings of the 2017 IEEE 60th International Midwest Symposium on Circuits and Systems (MWSCAS), Boston, MA, USA, 6–9 August 2017; pp. 1597–1600. [Google Scholar]
- Hu, Y.; Huber, A.; Anumula, J.; Liu, S.C. Overcoming the vanishing gradient problem in plain recurrent networks. arXiv 2018, arXiv:1801.06105. [Google Scholar]
- Polikar, R. Ensemble learning. In Ensemble Machine Learning; Springer: Berlin, Germany, 2012; pp. 1–34. [Google Scholar]
- Dietterich, T.G. Ensemble learning. Handb. Brain Theory Neural Netw. 2002, 2, 110–125. [Google Scholar]
- Zhang, C.; Ma, Y. Ensemble Machine Learning: Methods and Applications; Springer: Berlin, Germany, 2012. [Google Scholar]
- Brown, G. Ensemble Learning. Encycl. Mach. Learn. 2010, 6, 312. [Google Scholar]
- Pintelas, P.; Livieris, I.E. Special Issue on Ensemble Learning and Applications. Algorithms 2020, 13, 140. [Google Scholar] [CrossRef]
Feature | Description | Type | Values |
---|---|---|---|
Order | Pedogenic soil order | Nominal | {Alfisol, Inceptisol, Vertisol} |
Drainage | Soil drainage | Ordinal | {Very poorly, Poorly, Somewhat poorly, Moderately, Well, Very well} |
CaCO | Calcium carbonates | Ordinal | {None, Slight, Some, Strong} |
Texture | Soil texture in depth: 0–25 cm | Numerical (Real) | |
Texture | Soil texture in depth: 25–75 cm | Numerical (Real) | |
Texture | Soil texture in depth: 75–150 cm | Numerical (Real) | |
Slope | Percentage of slope | Ordinal | {Little, Moderate, High} |
Erosion | Soil erosion | Ordinal | {No erosion, Slightly, Moderately, Very, Severely} |
Feature | Description | Type | Values |
---|---|---|---|
Re-sowing | Re-sowing was or not performed | Nominal (Binary) | {True, False} |
Variety | Variety of seed | Nominal | {Acala (Zeta-0/Zeta-5), 4S} |
Seed | Kilograms of seed per 0.1 hectare | Numerical (Real) | |
Germination | Percentage of germination | Numerical (Percentage) | |
spacingOut | Spacing out | Categorical (Binary) | {True, False} |
XN | Kilograms of nitrogen per 0.1 hectare in oxide form | Numerical (Real) | |
XP | Kilograms of phosphorus per 0.1 hectare in oxide form | Numerical (Real) | |
Weeding | Weed control before sowing | Nominal (Binary) | {True, False} |
SPRAY | Weed control after emergence | Nominal (Binary) | {True, False} |
SPRAY | Control for Helicoverpa armigera | Numerical (Integer) | |
SPRAY | Insect control for aphids sp. | Numerical (Integer) | |
Irrger | Irrigation before germination | Numerical (Real) | |
Irrigat | Irrigation after emergence | Numerical (Real) |
Feature | Description | Type | Values |
---|---|---|---|
Defolation | If defoliation was performed. | Nominal (Binary) | {True, False} |
prodLoss | Loss of production due to extreme climatic conditions | Numerical (Percentage) | |
prevUse | Previous use of the farm | Nominal | {One year cereals, Two years cereals, One year cotton, Two years cotton, Three or four years cotton} |
typeHarv | Type of harvesting | Nominal | {By hand, Mechanized}. |
Days | Number of days between sowing and harvesting | Numerical (Integer) |
Model | Description |
---|---|
DTR | Splitting criterion: MSE, |
Max depth = 4, | |
. | |
GP | , |
. | |
FFNN | Architecture: 2 hidden layers with 50 and 10 neurons and an output layer of 1 neuron, |
Activation functions = ReLu, | |
Optimizer = Adam. | |
GRU | Architecture: 1 GRU layer with 40 units, 1 dense layer with 10 neurons and an output layer of 1 neuron, |
Activation functions = ReLu, | |
Optimizer = Adam. | |
LSTM | Architecture: 1 LSTM layer with 70 units, 1 dense layer with 20 neurons and an output layer of 1 neuron, |
Activation functions = ReLu, | |
Optimizer = Adam. | |
kNN | , |
Euclidean distance. | |
LASSO | , |
. | |
LR | No parameters specified. |
SVR | Kernel = RBF, |
, | |
, | |
, | |
. |
Model | Type | MAE | RMSE | MAPE | sMAPE |
---|---|---|---|---|---|
FFNN | Neural network-based regressors | 46.784 | 57.776 | 16.464 | 16.030 |
LSTM | 43.303 | 55.378 | 14.203 | 14.598 | |
GRU | 44.826 | 56.103 | 15.124 | 15.243 | |
DTR | State-of-the-art regressors | 45.942 | 57.920 | 17.081 | 15.407 |
GP | 41.475 | 54.634 | 15.697 | 13.964 | |
kNN | 45.441 | 54.963 | 16.030 | 15.236 | |
LASSO | 57.647 | 72.792 | 18.199 | 19.480 | |
LR | 41.480 | 54.641 | 15.699 | 13.965 | |
SVR | 65.517 | 79.045 | 20.024 | 22.881 | |
MNN | Proposed model | 40.479 | 52.741 | 14.272 | 13.632 |
MNN | 39.612 | 49.439 | 13.164 | 12.840 | |
MNN | 38.707 | 48.213 | 12.713 | 12.608 |
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |
© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
Share and Cite
Livieris, I.E.; Dafnis, S.D.; Papadopoulos, G.K.; Kalivas, D.P. A Multiple-Input Neural Network Model for Predicting Cotton Production Quantity: A Case Study. Algorithms 2020, 13, 273. https://doi.org/10.3390/a13110273
Livieris IE, Dafnis SD, Papadopoulos GK, Kalivas DP. A Multiple-Input Neural Network Model for Predicting Cotton Production Quantity: A Case Study. Algorithms. 2020; 13(11):273. https://doi.org/10.3390/a13110273
Chicago/Turabian StyleLivieris, Ioannis E., Spiros D. Dafnis, George K. Papadopoulos, and Dionissios P. Kalivas. 2020. "A Multiple-Input Neural Network Model for Predicting Cotton Production Quantity: A Case Study" Algorithms 13, no. 11: 273. https://doi.org/10.3390/a13110273
APA StyleLivieris, I. E., Dafnis, S. D., Papadopoulos, G. K., & Kalivas, D. P. (2020). A Multiple-Input Neural Network Model for Predicting Cotton Production Quantity: A Case Study. Algorithms, 13(11), 273. https://doi.org/10.3390/a13110273