Analysis of agriculture data using data mining techniques: application of big data
© The Author(s) 2017
Received: 25 February 2017
Accepted: 31 May 2017
Published: 5 July 2017
In agriculture sector where farmers and agribusinesses have to make innumerable decisions every day and intricate complexities involves the various factors influencing them. An essential issue for agricultural planning intention is the accurate yield estimation for the numerous crops involved in the planning. Data mining techniques are necessary approach for accomplishing practical and effective solutions for this problem. Agriculture has been an obvious target for big data. Environmental conditions, variability in soil, input levels, combinations and commodity prices have made it all the more relevant for farmers to use information and get help to make critical farming decisions. This paper focuses on the analysis of the agriculture data and finding optimal parameters to maximize the crop production using data mining techniques like PAM, CLARA, DBSCAN and Multiple Linear Regression. Mining the large amount of existing crop, soil and climatic data, and analysing new, non-experimental data optimizes the production and makes agriculture more resilient to climatic change.
It helps farmers in providing the historical crop yield record with a forecast reducing the risk management.
It helps the government in making crop insurance policies and policies for supply chain operation.
Data mining technique plays a vital role in the analysis of data. Data mining is the computing process of discovering patterns in large data sets involving methods at the intersection of artificial intelligence, machine learning, statistics, and database system. Unsupervised (clustering) and supervised (classifications) are two different types of learning methods in the data mining. Clustering is the process of examining a collection of “data points,” and grouping the data points into “clusters” according to some distance measure. The goal is that data points in the same cluster have a small distance from one another, while data points in different clusters are at a large distance from one another. Cluster analysis divides data into well-formed groups. Well-formed clusters should capture the “natural” structure of the data . This paper focuses on PAM, CLARA and DBSCAN clustering methods. These methods are used to categorize the different districts of Karnataka which are having similar crop production.
Clustering is considered as an unsupervised classification process . A large number of clustering algorithms have been developed for different purposes [4–6]. Clustering techniques can be categorised into Partitioning clustering, Hierarchical clustering, Density-based methods, Grid-based methods and Model based clustering methods.
Partitioning clustering algorithms, such as K-means, K-medoids PAM, CLARA and CLARANS assign objects into k (predefined cluster number) clusters, and iteratively reallocate objects to improve the quality of clustering results. Hierarchical clustering algorithms assign objects in tree structured clusters, i.e., a cluster can have data point’s representatives of low level clusters . The idea of Density-based clustering methods is that for each point of a cluster the neighbourhood of a given unit distance contains at least a minimum number of points, i.e. the density in the neighbourhood should reach some threshold. The idea of the density-based clustering algorithm is that, for each point of a cluster, the neighbourhood of a given unit distance has to contain at least a minimum number of points .
There are different forecasting methodologies developed and evaluated by the researchers all over the world in the field of agriculture. Some of such studies are: Researchers like Ramesh and Vishnu Vardhan are analysed the agriculture data for the years 1965–2009 in the district East Godavari of Andhra Pradesh, India. Rain fall data is clustered into 4 clusters by adopting the K means clustering method. Multiple linear regression (MLR) is the method used to model the linear relationship between a dependent variable and one or more independent variables. The dependent variable is rainfall and independent variables are year, area of sowing, production. Purpose of this work is to find suitable data models that achieve high accuracy and a high generality in terms of yield prediction capabilities .
Bangladesh offers several varieties of rice which has different cropping season . For this a prior study of climate (effect on temperature and rainfall) in Bangladesh and its effect on agricultural production of rice has been done. Then this study was being taken into regression analysis with temperature and rainfall. Temperature puts an adverse consequence on the crop production. The data has been taken from the “Bangladesh Agricultural Research Council (BARC)” for past 20 years with 7 attributes: “rainfall”, “max and min temperature”, “sunlight”, “speed of wind”, “humidity” and “cloud-coverage”. In Pre-processing, the whole dataset was divided in 3 month duration phases (March to June, July to October, November to February). For this duration, the average for every attribute has been taken and associated with it. This pre-processing has been done for each kind of rice variety. In clustering, the different pre-processed table has been analysed to find the sharable group of region based on similar weather attribute.
Soil characteristics are studied and analysed using data mining techniques. As an example, the k-means clustering is used for clustering soils in combination with GPS-based technologies . Authors like Alberto Gonzalez-Sanchez, Juan Frausto-Solis and Waldo Ojeda-Bustamante have done extensive study on predictive ability of machine learning techniques such as multiple linear regression, regression trees, artificial neural network, support vector regression and k-nearest neighbour for crop yield production . Wheat yield prediction using machine learning and advanced sensing techniques has done by Pantazi, DimitriosMoshou, Thomas Alexandridis and Abdul MounemMouazen . The aim of their work is to predict within field variation in wheat yield, based on on-line multi-layer soil data, and satellite imagery crop growth characteristics. Supervised self-organizing maps capable of handling existent information from different soil and crop sensors by utilizing an unsupervised learning algorithm were used. The software tool ‘Crop Advisor’ has been developed by S. Veenadhari, B. Misra and CD Singh  is an user friendly web page for predicting the influence of climatic parameters on the crop yields. C4.5 algorithm is used to find out the most influencing climatic parameter on the crop yields of selected crops in selected districts of Madhya Pradesh.
The objective of proposed work is to analyse the agriculture data using data mining techniques. In proposed work, agriculture data has been collected from following sources:
Crop wise agriculture data [html://CROPWISE_NORMAL_AREA],
Agriculture data based on weather, temperature, and relative humidity [http://dmc.kar.nic.in/trg.pdf].
Input dataset consist of 6 year data with following parameters namely: year, State-Karnataka (28 districts), District, crop (cotton, groundnut, jowar, rice and wheat.), season (kharif, rabi, summer), area (in hectares), production (in tonnes), average temperature (°C), average rainfall (mm), soil, PH value, soil type, major fertilizers, nitrogen (kg/Ha), phosphorus (Kg/Ha),Potassium(Kg/Ha), minimum rainfall required, minimum temperature required.
In proposed work, modified approach of DBSCAN method is used to cluster the data based on districts which are having similar temperature, rain fall and soil type. PAM and CLARA are used to cluster the data based on the districts which are producing maximum crop production (In proposed work wheat crop is considered as example). Based on these analyses we are obtaining the optimal parameters to produce the maximum crop production. Multiple linear regression method is used to forecast the annual crop yield.
Modified approach of DBSCAN
Modified DBSCAN proposes the method to find the minimum points and Epsilon (radius value) automatically. KNN plot is used to find out the epsilon value where input to the KNN plot (K value) is user defined. To avoid the user define K value as input to the KNN plot, Batchelor Wilkins clustering algorithm is applied to the database and obtain the K value along with its respective cluster centres. This K value is given as input to the KNN Plot.
Determination of Eps and Minpts
The Epsilon (Eps) value can be found by drawing a “K-distance graph” for entire data-points in dataset for a given ‘K’, obtained by the Batchelor Wilkins Algorithm . Initially, the distance of a point to every ‘K’ of its nearest-neighbours is calculated. KNN plot is plotted by taking the sorted values of average distance values. When the graph is plotted, a knee point is determined in order to find the optimal Eps value .
Partition around medoids (PAM)
BUILD phase, a collection of k objects are selected for an initial set S.
Arbitrarily choose k objects as the initial medoids.
Until no change, do.
(Re) assign each object to the cluster with the nearest medoid.
Improve the quality of the k-medoids (randomly select a non medoid object, O random, compute the total cost of swapping a medoid with O random).
SWAP phase, one tries to improve the quality of the clustering by exchanging selected objects with unselected objects. Choose the minimum swapping cost.
CLARA (clustering large applications)
Multiple linear regression to forecast the crop yield
Before applying the multiple linear regression to forecast the crop yield, it’s necessary to know the significant attributes from the database. All the attributes used in the database will not be significant or changing the value of these attributes will not affect anything on the dependent variables. Such attributes can be neglected. P value test is performed on the database to find the significant attributes and multiple linear regression is applied only on the significant values to forecast the crop yield.
Data mining algorithms work with different principles, being able to be influenced by different kinds of associations on data. To ensure fairer conditions in evaluation, this work finds the optimal clustering method for agriculture data analysis. Proposed work adopts the external quality metrics  like Purity, Homogeneity, Completeness, V Measure, Rand Index, Precision, Recall and F measure to compare the PAM, CLARA and DBSCAN clustering methods.
Purity of the clustering is computed by assigning each cluster to the class which is most frequent in the cluster. Homogeneity represents the each cluster contains only members of a single class. Completeness represents the all members of a given class are assigned to the same cluster. V-measure is computed as the harmonic mean of distinct homogeneity and completeness scores. Rand Index measures the percentage of decisions that are correct. Precision is calculated as the fraction of pairs correctly put in the same cluster. Recall represents the fraction of actual pairs that were identified. F measure indicates the harmonic mean of precision and recall. Higher quality metrics value represents the better cluster quality.
Modified approach of DBSCAN
KNN plot is plotted using K value obtained from the Batchelor & Wilkins’ Algorithm to determine the epsilon value and the min points for the DBSCAN.
DBSCAN clustering algorithm is applied on the dataset to cluster the different districts of Karnataka which are having similar rain fall, temperature and soil type using optimal Eps value.
Figure 6 depicts the different districts of Karnataka which are considered for the purpose of analysis.
Study and analysis of wheat crop production in different districts of Karnataka as shown in Fig. 10.
Results of PAM algorithm
Mandya, Raichur, Gadag, Gulbarga, Bellary
Koppal, Dharwad, Haveri, Bijapur, Bidar, Chamarajannagar, Belgaum, Tumkur
Davangere, Shimoga, Chikmagalur, Bangalore
As a result of the analysis, North Karnataka districts such as Bijapur, Dharwad, Bagalkot, Belgaum, Raichur, Bellary, Chitradurga and Davangere are the districts which have maximum wheat crop production.
Results CLARA algorithm
Large area, production and moderate rainfall, temperature (24–26)
Moderate area, production and high rainfall, temperature (27–29)
Low area, production moderate rainfall, temperature (29–30)
Gadag, Gulbarga, Dharwad, Bangalore, Bagalkote
Koppal, Davangere, Shimoga, Haveri, Chikmagalur, Bidar, Chamarajannagar, Tumkur, Mandya, Raichur, Bellary
Multiple linear regression
P value test: significant attributes
Multiple linear regression equation for different crop yield
Yield forecast equation
Yield = (7.149372) + (−0.14468)pH + (−0.00131) Nitrogen + (−0.00405) Potassium + (−0.00405) Water Required
Yield = (2.79115) + (0.029217) Temperature + (5.78e–05) Rainfall + (−0.05681) pH + (−0.00127) Phosphorus + (−0.00492) Water Required
Yield = (−1.62694) + (−5.35e–02) Temperature + (0.051512) pH + (−0.00113) Potassium + (0.01685436) Water Required
Yield = (−0.18503) + (0.041593) Temperature + (0.172042) pH + (−8.27e–04) Nitrogen + (−4.28e–03) Phosphorus + (−0.00264) Potassium + (9.15e–04) Water Required
Yield = (112) + (−4.14e–02) Temperature + (1.34e–04) Rainfall + (0.079153) pH + (−1.31e–03) Nitrogen + (−0.00167) Potassium + (−0.28125) Water Required
For 1 unit increase in pH, the crops like Jowar, Rice, and Wheat yield will increase but Groundnut and Cotton yield will decrease.
Results for optimal temperature and rainfall for wheat—Table 5
Optimal parameters to achieve higher production
Optimal parameters to achieve higher production
Comparison of clustering methods
Comparison of clustering methods
Number of clusters k = 3
The crops are usually selected by its economic importance. However, the agricultural planning process requires a yield estimation of several crops. In this sense, five crops were selected for this work using the data availability as the key measure. Thus, a crop was selected when enough data samples appeared in the range of 6 years under analysis. In presents works, research is commonly limited to the 5 crops those are cotton, wheat, ground nut, jowar and rice. Example wheat crop analysis is discussed in this paper.
The present work covers the PAM, CLARA, Modified DBSCAN clustering methods and multiple linear regression method. PAM and CLARA are the traditional clustering methods where as DBSCAN method is modified by introducing the Batchelor Wilkins clustering method to determine the ‘k’ value and KNN method to determine the minimum points and radius value automatically. Using these methods crop data set is analysed and determined the optimal parameters for the wheat crop production. Multiple linear regression is used to find the significant attributes and form the equation for the yield prediction.
Some works measure the quality of the clustering methods using internal quality metrics , some other uses the external quality metrics. However, in these works, research is limited to the external quality metrics which are combination of several metrics those are : set matching metrics, metrics based on counting pairs and metrics based on Entropy. The quality metrics were ranked, from the best to the worst, according to purity, homogeneity, completeness, v measure, precision, recall and rand index results, in the following order: DBSCAN, CLARA and PAM.
Various data mining techniques are implemented on the input data to assess the best performance yielding method. The present work used data mining techniques PAM, CLARA and DBSCAN to obtain the optimal climate requirement of wheat like optimal range of best temperature, worst temperature and rain fall to achieve higher production of wheat crop. Clustering methods are compared using quality metrics. According to the analyses of clustering quality metrics, DBSCAN gives the better clustering quality than PAM and CLARA, CLARA gives the better clustering quality than the PAM. The proposed work can also be extended to analyse the soil and other factors for the crop and to increase the crop production under the different climatic conditions.
JM, Dean R&D, Prof & HOD of Dept of M.Tech CSE at NMIT, has 40 years of experience in India and abroad has guided and given extensive help to develop the data mining algorithms. SN, Assistant Professor of Dept of M.Tech CSE at NMIT has developed the PAM and CLARA algorithms with the help of Dr. Jharna Majumdar. SA Assistant Professor of Dept of M.Tech CSE at NMIT has developed Modified approach of DBSCAN, Multiple Linear Regression and quality metrics for cluster comparison with the guidance and help of Dr. Jharna Majumdar. All authors together analysed the crop data set to determine the optimal parameters to maximise the crop yield. All authors read and approved the final manuscript.
The authors express their sincere gratitude to Prof N.R Shetty, Advisor and Dr H.C Nagaraj, Principal, Nitte Meenakshi Institute of Technology for giving constant encouragement and support to carry out research at NMIT.
The authors extend their thanks to Vision Group on Science and Technology (VGST), Government of Karnataka to acknowledge our research and providing financial support to setup the infrastructure required to carry out the research.
The authors declare that they have no competing interests.
This work was supported by the Research Department of Computer science, Nitte Meenakshi Institute of Technology.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
- Veenadhari S, Misra B, Singh CD. Data mining techniques for predicting crop productivity—A review article. In: IJCST. 2011; 2(1).Google Scholar
- Gleaso CP. Large area yield estimation/forecasting using plant process models.paper presentation at the winter meeting American society of agricultural engineers palmer house, Chicago, Illinois. 1982; 14–17Google Scholar
- Majumdar J, Ankalaki S. Comparison of clustering algorithms using quality metrics with invariant features extracted from plant leaves. In: Paper presented at international conference on computational science and engineering. 2016.Google Scholar
- Jain A, Murty MN, Flynn PJ. Data clustering: a review. ACM Comput Surv. 1999;31(3):264–323.View ArticleGoogle Scholar
- Jain AK, Dubes RC. Algorithms for clustering data. New Jersey: Prentice Hall; 1988.MATHGoogle Scholar
- Berkhin P. A survey of clustering data mining technique. In: Kogan J, Nicholas C, Teboulle M, editors. Grouping multidimensional data. Berlin: Springer; 2006. p. 25–72.View ArticleGoogle Scholar
- Han J, Kamber M. Data mining: concepts and techniques. Massachusetts: Morgan Kaufmann Publishers; 2001.MATHGoogle Scholar
- Ester M, Kriegel HP, Sander J, Xu X. A density-based algorithm for discovering clusters in large spatial databases with noise. In: Paper presented at International conference on knowledge discovery and data mining. 1996Google Scholar
- Ramesh D, Vishnu Vardhan B. Data mining techniques and applications to agricultural yield data. In: International journal of advanced research in computer and communication engineering. 2013; 2(9).Google Scholar
- MotiurRahman M, Haq N, Rahman RM. Application of data mining tools for rice yield prediction on clustered regions of Bangladesh. IEEE. 2014;2014:8–13.Google Scholar
- Verheyen K, Adrianens M, Hermy S Deckers. High resolution continuous soil classification using morphological soil profile descriptions. Geoderma. 2001;101:31–48.View ArticleGoogle Scholar
- Gonzalez-Sanchez Alberto, Frausto-Solis Juan, Ojeda-Bustamante W. Predictive ability of machine learning methods for massive crop yield prediction. Span J Agric Res. 2014;12(2):313–28.View ArticleGoogle Scholar
- Pantazi XE, Moshou D, Alexandridis T, Mouazen AM. Wheat yield prediction using machine learning and advanced sensing techniques. Comput Electron Agric. 2016;121:57–65.View ArticleGoogle Scholar
- Veenadhari S, Misra B, Singh D. Machine learning approach for forecasting crop yield based on climatic parameters. In: Paper presented at international conference on computer communication and informatics (ICCCI-2014), Coimbatore. 2014.Google Scholar
- Rahmah N, Sitanggang IS. Determination of optimal epsilon (Eps) value on DBSCAN algorithm to clustering data on peatland hotspots in Sumatra. IOP conference series: earth and environmental. Science. 2016;31:012012.Google Scholar
- Forbes G. The automatic detection of patterns in people’s movements. Dissertation, University of Cape Town. 2002.Google Scholar
- Ng RT, Han J. CLARANS: A Method for Clustering Objects for Spatial Data Mining. In: IEEE Transactions on Knowledge and Data Engineering. 2002; 14(5).Google Scholar
- Kaufman L, Rousseeuw PJ. Finding groups in data: an introduction to cluster analysis. Wiley. 1990. doi:10.1002/9780470316801.MATHGoogle Scholar
- Multiple linear regression-http://www.originlab.com/doc/Origin-Help/Multi-Regression-Algorithm. Accessed 3 July 2017.
- Elbatta MNT. An improvement for DBSCAN algorithm for best results in varied densities. Dissertation, Gaza (PS): Islamic University of Gaza. 2012Google Scholar
- Kirkl O, De La Iglesia B. Experimental evaluation of cluster quality measures. 2013. 978-1-4799-1568-2/13. IEEE.Google Scholar
- Meila M (2003) Comparing clustering. In: Proceedings of COLT 2003.Google Scholar