Machine learning project

Predicting Fertility of Soil Using Data Mining
Techniques
School of Computing
National College of Ireland
Nikhil Deshmukh, Shailesh Rathi, Shital Thakare, Siddharth Chaudhary
{x16145003, x16138163, x16139739, x16137001}@student.ncirl.ie
Abstract— Agriculture industry in India is the greatest
sector for employment but lack of research in this sector is the
reason behind less productivity. It's important to implement
computational research, Machine learning techniques in
Agriculture industry to make India better quality and quantity
producer in food sector. Machine Learning techniques are useful
in abstracting patterns and establishing relationships between
varied data sets and predicting reasonable outputs. It can be
efficiently applied in Agriculture industry to improve efficiency
in this sector. We have discussed application of Machine learning
techniques in Agriculture sector to analyze fertility of soil.
Agriculture industry has been always one of the interested area of
research. This study venture to analyze soil data depending
upon various factors, classify it and improve efficiency of each
model using different combinations.
Keywords— ANN; SVM; Decision Tree; KNN; Soil;
I. INTRODUCTION
Agriculture is one of the highest employer sector in India as
well as worldwide. Around 51% population in India has been
employed by Agriculture industry. Contradictory to that it is
accountable for only 18% of annual GDP. The mismatch is
due to lack of research and less use of technologies.
Agriculture sector in India is less automated as compared
to western countries. Western countries using various
technology to do predictions in Agriculture. Indian
agriculture sector is lacking behind in this area. Productivity
and growth is very much less due to lack of technology.
We have considered Indian soil data to predict the fertility
of soil considering various elements like nitrogen, oxygen,
Phosphorous, porosity and various environmental elements like
air, temperature. Data has been taken from ISRIC -World soil
information [1]. ISRIC is an institute which works on soil
science. India has diverse weather conditions as well as soil
type across different regions. Data set belongs to Haryana
state in India which ground of moist soil.
fertility of soil the limiting factor in Agricultural industry
India. Soil fertility defines growth of plants when other
environmental factors like light, water, temperature are
favorable. Soil fertility is influence by several factors like
Climate, irrigation (Soil water), Soil, acidity, Soil alkalinity,
Nutrition in Soil. Globalization, changing weather condition,
urbanization, higher use of pesticides is the reason of
decreasing quality of soil in India. Deficient soil type lead to less
agricultural production and ultimately higher cost of food
products. Different soil types are used to analyze fertility
of soil. The ultimate goal of applying technology in Agricultural
with minimal impact in fertility of soil and quality of food
product [4].
Machine Learning techniques are useful in abstracting patterns
and establishing relationships between varied data sets and
predicting reasonable outputs. Agriculture industry in India is the
greatest sector considered for employment and has been part
of research. Machine learning techniques can be efficiently
applied in Agriculture industry to improve research. In current
scope of project, we have developed a model for fertility of
soil based on different soil type. After receiving fertility
depending upon various soil type, a comparative study of
machine learning techniques such as ANN, linear
regression, SVM and Decision tree is carried out.
II. RELATED WORK
Agriculture is one of the hot topic of research among
academic as well as scientific researchers. A lot of previous
research has been done in Agriculture industry and fertility of
soil by using different data mining, classifications and
statistical techniques.
Various factors affect fertility of soil. Among which
Sulphur, Water and Zinc are the most influencing factors
constituting fertility of soil. A agriculture case study carried
out on 3622 soil samples from different districts in India. It is
concluded in study that water or moisture level in soil is the
most influencing factor. Soil fertility varies depending
moisture level in soil[15].
Author[1]used segmentation algorithm to divide signals
and features. Signals are extracted using boundary method and
then classifiers divided into classes; they used SVM, Decision
Trees, and ANN to classify surface soil data.
Author[2]developed hierarchical neural network models to
predict water retention and hydraulic conductivity.
Author[3]developed regression and artificial neural network to
check the water retention with the help of texture and bulk
density. Ahmad [4] predicted soil moisture using SVM for
data sensing remotely. Author[6] implemented linear
regression technique for the forecasting of soil data along with

Naïve Bayes, J48 classification. Armstrong [5] implemented
cluster analysis on soil data collected by the food department of
Australia.
Author[8] used different classification techniques on soil
texture and found Bayesian classification is more accurate and
performance is also good. [9]Gholap did the comparative
analysis between classification techniques; such as Naïve
Bayes, J48 and classical linear regression and found least
median regression produced better result. Author
[10]implemented artificial neural network and digital terrain-
analysis for high quality soil maps. Author[12] Foody
implemented single Support Vector Machine against series of
the classifier for crop classification.
[13]implemented decision tree to the fields to help the
farmers in making the decision to select pump for the
irrigation and it’s depend on irrigation types, total area
coverage of the field, capacity of the motor, and the height.
Several techniques applied different classifications machine
learning methods, such as J48, Naïve Bayes, and Random
forest algorithm to classify the fertility of the soil. J48 gives
better result than other algorithms. Rub implemented multiple
regression techniques on soli data and concluded SVM
generated better model for the predication.
III. METHODOLOGY
Cross Industry Standard for Data Mining Model (CRISP-DM)
has been used among various methodologies for our data
mining project as it best suits our business due to its extensive
step by step development ability. This is the six-phase process
which will be covered in the remaining sections.
[CHAPMAN]
Fig 1: Understanding the business
A. Understanding the Bisiness : This is the first step in the
process of data mining, before proceeding to the next step
all the informations has gathered related to the Agriculture
and soil business.
Business Objectives and Business Questions are as follows:
1. Which properties are important in defining the
fertility of soil?
2. Which properties of soil are highly correlated to each
other?
3. Which properties do not vary with respect to fertility
of soil?
4. Which Classifier performs best for predicion of soil
data?
B. Understanding the soil data : Raw data set collected from
the ISRC (International Soil Reference and Information
Centre) an organization which works in soil research area
as well as different types of soils across world. Data set
selected had 60 attributes, relevant attribute as per
research has been selected. Final dataset contains 10
different attributes and around 1300 records. Attributes
arewhich are as follows:
Table1: Data set details
C. Data Preperation :
. Data cleaning and formatting needed before using it as final
input. Removed unnecessary columns, null values, extra blank

spaces in dataset using R programming language and basic
excel functionality. Some packages like readr, tidyr have been
used.
Following step has been performed in data cleaning process.
1. Null Values handling: There were very few null records in
dataset. Null records replaced with NA values to remove
inconsistency in the data set.
2. Removing irrelevant columns: Data set contain some
columns which were irrelevant for our research, removed
those unnecessary columns, only relevant data has been taken as
input for processing.
3. Inconsistent Data Types: Numeric values in data set were
not in consistent format ex. Precision and scale defined for
'Porosity' column was not consistent, unique precision
modified for this column.
4.Unnessary Spaces handling:
Used 'TRIM' function in excel to remove unnecessary spaces in
column values.
Processed clean data is used as input for further
analysis. D. Data Modeling and Evaluation
Sequence for data modeling and evaluation is selecting the
technique, generate, test design, building a model, model
assessment, evaluation of result, Review and determine the
next step [ASTESJ]
1) Decision Tree
QUESTION 1: Which properties are important in defining
the fertility of soil?
Fig : 2 Decision tree model
In our project, we used decision tree as classification model.
There are three labels in the class in dataset depending upon
fertility of soil which are High Medium and low. Various
dependent factors considered are ~ ph,depth, conductivity
,carbon ,Nitrogen, Phosphorus, Potassium, WHC, Porosity.
Once data is loaded, shuffling is done on dataframe. Shuffled
dataframe is then divided into train and test, 70:30 percent
ratio is maintained for train and test. Decision tree is applied on
factors mentioned dependent factors mentioned above. R part
function is used to create decision tree and classes are
predicted depending variables. A prediction model is prepared
using decision tree as input on test data set. Graph is plotted
using prediction model.
Fig :3 Decision tree for soil data
Table 2: Confusion matrix for DT
Accuracy= sum(diag(table))/sum(table)
accuracy = (16 + 90 + 87)/ (16 + 4 + 17 + 2 + 90 + 24 + 30 +
34 + 87)
accuracy = 193 / 304 = 0.63486
accuracy = 63.48 %
ANSWER 1: To answer this question we have applied
decision tree classifier on the soil data, decision tree evaluates
the entropy of each attribute to decide the root of the tree.
Decision tree classification is shown in the figure, the root of the
tree is Conductivity, which we can assume is the most
important criteria, next one will be the water holding capacity.

Below is the list of factors which are arrange in order of
significance of role in classification
1) Conductivity
2) WHC
3) Potassium
4) Nitrogen
5) pH
6) Phosphorous
2) ANN
QUESTION 2: Which properties of soil are highly
correlated to each other?
We have used ANN as it gives good result for classification
[5]. For ANN, we have changed the class as numeric as ANN
does not take the string as an input. We have used different
hidden node to compare our result 1, 3, 5 and 7 respectively.
While implementing ANN first step was to load the complete
soil data in R data frame. In this, classes were in the form of
string High, mid and low fertile which was changed into 3,2,1
respectively. Due to different Range of ever column, data has
normalized. Using the min max formula. Data has been split
into train and test with 80:20 split. Data has been trained using
neuralnet function by using neuralnet package for different
hidden nodes. Using cor function percentage of corret
classified data in ANN has been identified between class and
type of the soil.
Accuracy and error can be seen in below table:
ANN nodes Training RMS Prediction
Steps percentage
ANN with 1 node 2085 31.59 48
ANN with 3 34672 19.22 50
nodes
ANN with 5 42683 15.42 52
nodes
ANN with 7 97794 13.92 55
nodes
Table 3 Accuracy and error table for ANN
Fig4 : ANN with single hidden node
Fig 5: ANN with 7 hidden nodes

Fig : 6 Matrix correlation
ANSWER 2: To answer this question we have performed
correlation analysis with the help of R Fig 6 Matrix correlation
represents the relationship between the entire soil data as we can
see from Fig 6, the highest correlation of 0.81 is between WHC
and porosity. Next 2 properties which are highly correlate
are Carbon and Nitrogen with coefficient value of
0.79, next one is conductivity and pH. The upper triangular
matrix of this fig shows the correlation coefficient and lower
triangular matrix is consisting of scatter plot between each pair
of variable.
3) SVM
For support vector machine, we do not need to convert the
final string class into numeric data. Our Class label contain 3
labels: low fertile, med fertile and high fertile.
Support vector machine is one of the most powerful technique in
machine learning. SVM combines both the concepts:
clustering as well as regression. SVM is a black box technique
which is generally used for prediction and classification
problems. SVM can be thought as which creates a two-
dimension boundary on a surface between different data points
to form two different class. The decision boundary should be
equidistant from both the labels of the class. Support vectors
validate the distant of a hyperplane.as shown in Fig 7. This
dataset can’t be classified using simple SVM. We have used
different kernels like Polynomial, Radial Basis, Hyperbolic
Tangent.
Fig 7: SVM hyper plane
Here is the description of the kernel used and their
performance. Table 4 radial basis Kernel gives best
performance.
Kernel type Prediction Support Training
percentage Vectors error
Polynomial 69.00 767 .32
Radial Basis 80.00 714 .15
Hyperbolic 44.00 751 .62
tangents
Table 4: Description of SVM result for different Kernels
Technique 2 for SVM: For SVM, we have also used the
technique of one vs all for classification and kernels like
polydot, vanilladot, rbfdot, splinedot for this dataset which are
given in the code. For each label for the class accuracy is
calculated. Splinedot gives the best accuracy rate. Below is the
accuracy, precision, recall, F-measure of the SVM using
splinedot for 3 different labels of the class. we have used
f- measure to calculate the performance.
Label Accuracy Precision Recall F-
Class measure
High 87% 66% 78% 71.5%
fertile
Med 82% 76% 76% 76%
fertile
Low 93% 83% 85% 83.9%
fertile
Table 5: Accuracy, precision, recall and F-measure for SVM

4) KNN
QUESTION 3: Which properties do not vary with respect to
fertility of soil?
After applying all the classification methods on Soil data
ANN, SVM and DT, then we applied kNN classification
method on class labeled type, to check classification accuracy of
the algorithm. In this algorithm, we used on the labeled
class, which contains three categorical values.
In the data, values of the variables are different, so we
used normalization technique to transform the data into
common scale. After that, data divided into two sets training and
testing with 70 :30 ratio. And then used kNN function on the
dataset with the value K= 17, which played significant
role to determine the efficiency of the model, and value
calculated by the square root of the test observation, and plot the
confusion matrix on the class label to check the accuracy, Kappa
statics. And plotted the histogram to check the accuracy of the
dataset.
From the Table 6 we can state that Pottasium, Carbin and pH are
not statistically significant. So, WHC, Nitrogen and
Phosphorous does not vary with respect to the fertility o soil.
QUESTION 4: Which Classifier performs best for
prediction of soil data?
ANSWER 4: In this question, we have applied four
important techniques of classification which are decision tree,
ANN, SVM, kNN. For each algorithm data is divided in test
and train. The performance of classifier is compared on fact that
how each of them has classified data correctly.
Techniques Accuracy
Decision Tree 63.48 %
ANN 55 % SVM
80%
kNN 70%
Table 7: Comparison of accuracy
From the above Table 7 we can conclude that SVM performs
best among all the technique.
5) Deployment
This is 6th
and final phase of the life cycle in which accuracy
is monitored for all the results and final report has been
prepared.
Fig:8 Frequency of soil fertility
ANSWER 3: To answer this question we have performed
ANOVA of each property according to soil type in R and
results are as follows
IV. CONCLUSION AND FUTURE WORK
In this paper, we have implemented and executed different
classification methods such as Decision Tree, ANN, SVM and
kNN. The result of SVM out performs among all the
techniques. CRISP-DM methodology has been used with the
help of R tool.
In Future, we can collect more data from different parts of the
country and soil recommendation system can be built for the
commercial use which can help grow agriculture industry.
Property F-Statistic P Value
Potassium 0.42 0.51
WHC 352.88 2.2e-16***
Phosphorous 16.387 5.473e-05***
Nitrogen 30.074 5.004e-08***
Carbon 8.5978 0.003
pH 3.84 0.0502
Table 6: ANOVA output
References
[1] Bhattacharya, B., & Solomatine, D. P. (2006). Machine learning in soil
classification. Neural Networks, 19(2), 186-195.
[2] Schaap, M. G., Leij, F. J., & Van Genuchten, M. T. (1998). Neural
network analysis for hierarchical prediction of soil. hydraulic properties.
Soil Science Society of America Journal, 62(4), 847-855. [565]

[3] Pachepsky, Y. A., Timlin, D., & Varallyay, G. Y.(1996).Artificial neural
networks to estimate soil water retention from easily measurable data.
Soil Science Society of America Journal, 60(3), 727-733. [320]
[4] Ahmad, S., Kalra, A., & Stephen, H. (2010). Estimating soil moisture
using remote sensing data: A machine learning approach.Advances
in Water Resources, 33(1), 69-80.
[5] Armstrong, L. J., Diepeveen, D., & Maddern, R. (2007,December). The
application of data mining techniques to characterize agricultural soil
profiles. In Proceedings of the sixth Australasian conference on Data
mining and Analytics-Volume 70 (pp. 85100). Australian Computer
Society, Inc.
[6] Baskar, S. S., Arockiam, L., & Charles, S. (2013). Applying data mining
techniques on soil fertility prediction. International Journal of Computer
Applications Technology and Research, 2(6), 660-meta..
[7] Chandrakar, P. K., Kumar, S., & Mukherjee, D. (2011). Applying
classification techniques in Data Mining in agricultural land soil.
International Journal of Computer Engineering, 2, 89-95.
[8] Gholap, J., Ingole, A., Gohil, J., Gargade, S., & Attar, V (2012).
Soil data analysis using classification techniques and soil attribute
prediction. arXiv preprint arXiv:1206.1557
[9] Behrens, T., Förster, H., Scholten, T., Steinrücken, U.,Spies, E. D.,
& Goldschmitt, M. (2005). Digital soil mapping using artificial neural
networks. Journal of plant nutrition and soil science, 168(1),21-33.
[10] Paul, M., Vishwakarma, S. K., & Verma, A. (2015,December). Analysis
of Soil Behavior and Prediction of Crop Yield Using Data Mining
Approach. In Computational Intelligence and Communication Networks
(CICN), 2015 International Conference on (pp. 766-771). IEEE.
[11] Foody, G. M., & Mathur, A. (2004). A relative evaluation of multiclass
image classification by support vector machines. IEEE Transactions on
geoscience and remote sensing, 42(6), 1335-1343.
[12] Ravindra, M., Lokesha, V., Kumara, P., & Ranjan, A.
Study and Analysis of Decision Tree Based Irrigation
Methods in Agriculture System.
[13] Herman,Robinson T,Giovanni P,Alvaro N. A comparative between
CRISP-DM and SEMMA through the construction of a MODIS
repository for studies of land use and cover change.
[14] Chapman P, Clinton J, Kerber R, Khabaza T, Reinartz T, Shearer C,
Wirth R. CRISP-DM 1.0: Step-by-step data mining guide, 2000.

Machine learning project

More Related Content

What's hot (14)

Similar to Machine learning project (20)

More from Siddharth Chaudhary (19)

Recently uploaded (20)

Machine learning project