=Paper=
{{Paper
|id=Vol-1755/40-45
|storemode=property
|title=A Model for Prediction of Kidney Cancer Using Data Analytic Technique
|pdfUrl=https://ceur-ws.org/Vol-1755/40-45.pdf
|volume=Vol-1755
|authors=Felix Aranuwa,Olanike Ogundare,Sellappan Palaniappan
|dblpUrl=https://dblp.org/rec/conf/cori/AranuwaOP16
}}
==A Model for Prediction of Kidney Cancer Using Data Analytic Technique==
https://ceur-ws.org/Vol-1755/40-45.pdf
An Efficient Algorithm for the Prediction of Cancer of the
Kidney Using Data Analytic Technique
Aranuwa Felix Ola Ogundare Olanike Sellappan Palaniappan
Aekunle Ajasin University, Malaysia University of Science and Technology, Malaysia University of Science and Technology,
Akungba – Akoko, Ondo State, Nigeria Selangor, Malaysia Selangor, Malaysia +60192600962
+2347031341911 +6010212624
sell@must.edu.my
felix.aranuwa@aaua.edu.ng ogundareolanike@yahoo.com
ABSTRACT diagnosed at an advanced stage of the disease which usually
Our focus in this research work is to present an efficient algorithm contributes to its complications and mortality rate. This is due to a
for apt prediction of cancer of the kidney in which medical limited awareness of the early signs and symptoms of the disease
practitioners and patients could gain valuable knowledge for early among the public and healthcare providers. According to
and proactive intervention strategies to save lives from this Lasebikan, Nwadinigwe & Onyegbule, (2014), the mortality rates
harmful disease. To achieve these objectives, dataset pertaining to of this disease is always compounded by the later stage at which
patients of cancer of the kidney were acquired from selected the disease is diagnosed, presenting a ticking time bomb of life
private and public hospitals in south west Nigeria. A two-layered expectancy and lifestyle changes such as women having fewer
classifier system consisting of Rule Induction (RI) and Decision children, as well as hormonal intervention such as post-
Tree (DT) classifiers was designed to build the model based on menopausal hormonal therapy [1]. To reduce this harm caused by
data analytic approach. The classifier system designed was tested the disease, an effective way is to detect it early [2]. However,
successfully using case study data from fifty-two (52) selected early detection and prognosis requires an accurate information,
Local Governments in South West Nigeria using purposive and reliable analytic procedure and efficient algorithm. Therefore, the
selective sampling technique. Ten classification algorithms were researcher’s direction in this work is to present a reliable analytic
used in the modeling. Waikato Environment for Knowledge procedure and efficient algorithm suitable for the prediction of
Analysis was used for the experiment and each model was built in cancer of the kidney through data analytic approach, in which
two different ways (10-fold cross validation and percentage split medical practitioners and patients can gain valuable knowledge
mode). Performance comparison of the various algorithms and help for proactive intervention strategies in order to save lives
considered was carried out using standard metrics of accuracy for from this harmful disease.
classification and speed of model building benchmarks. The
experimental results show that the J48 decision tree algorithm Data analytic has proven to be a multi-dimensional discipline that
outperform all other algorithms in all the layers with correctly uses descriptive techniques and predictive models to gain valuable
classified instances of 74.7%, F-Measure of 0.614, TP rate of knowledge from data warehouses for recommendations and
0.747, FP rate of 0.135, precision and recall of 0.687 and 0.714 decision making. It is the discovery of patterns and
respectively. It took the best algorithm, 0.03 seconds to build the communication of meaningful insight in data [3]. According to
model. This proves that the algorithm is suitable for the research Berson, Smith and Thearling (1999), data analytics is the science
purpose. The results from the system framework when tested with of examining raw data with the purpose of drawing conclusions
test data shows that the identified attributes, algorithm and the from it [9]. It focuses on inference, identify undiscovered patterns
system model performed well and can serve as valuable tool for and establish hidden relationships[4]. Figure 1 depicts the process
early detection of the disease in patients. of data analytics. The science is generally divided into exploratory
data analysis (EDA), where new features in the data are
CCS Concepts discovered and confirmatory data analysis (CDA) where existing
• Software and its engineering ➝Software organization and
hypotheses are proven true or false. Typically, it is used to
properties ➝Extra-functional properties ➝Software
describe the technical aspects of data analysis,
performance
especially predictive modeling, machine learning techniques. Data
Keywords Analytics has been commonly apply to business data, marketing
Data Analytics, Classification Algorithms, Data Mining, Kidney mix modeling, web analysis, risk analysis and fraud analysis to
Cancer communicate insights from data. It is very good in recommending
action and guide decision making,
1. INTRODUCTION
In Africa, experimental studies have shown that most cancers are
CoRI’16, Sept 7–9, 2016, Ibadan, Nigeria.
Figure 1: Data Analytics Process
40
7 Age Group family history of kidney cancer; having kidney disease that
needs dialysis; being infected with hepatitis C; and previous
20-30 38 3.8 treatment for testicular cancer or cervical cancer. There is an
31-40 150 15.0 indication also, that High blood pressure is a possible risk factor
though still under investigation.
41-50 231 23.0
51-60 240 23.9 Table 2: Statistical Data for the Selected Attributes
61-70 211 21.0 S/N Attributes Data Percentage
70 -80 94 9.13 (%)
1 Gender
81-90 42 4.17
Male 451 44.8
91-100 0 0 Female 556 55.2
2. METHOD AND MATERIALS 2 Lifestyle
S/N Variable Name Variable Format Variable Type Smoking 397 39.5
1 Gender Male, Female Categorical Obesity 19 1.9
2 Age 25, 30,…….. Numerical
Drug Abuse 134 13.3
3 Lifestyle Smoking, Obesity, Categorical
4 G&H Disorder Yes, No Categorical HB Pressure 106 10.53
5 C & I Exposure Yes No Categorical Water Pills 40 3.98
6 Prediction One, Two, Three Categorical Dialysis 8 0.8
Level
Alcohol 295 29.3
2.1 Data Collection and Data Format
Dataset pertaining to this research work was collected from Radiation 7 0.69
selected health centres and hospitals in the south western part of 3 G&H Disorder
Nigeria using purposive and selective sampling techniques. The
researcher collected a sample data totaling, 1,006 records from Yes 329 32.7
fifty-two selected health centres in six (6) different states. The No 677 67.3
data collected was cleaned, normalized and organized in a form
4 C&I Exposure
suitable for data analytic process. Table 1 shows the data format
for the research data collection while Figure 1 and Figure 2 show Yes 576 57.3
the visualized information about selected states and health centres
No 430 42.7
respectively.
5 Complaints
Table 1 shows the data format for the research data collection Blood in Urine 113 11.23
Back pain 203 20.17
Tumor 189 18.8
Fibroid 131 13.02
6 Stomach Ucher 144 14.31
Kidney pain 159 15.8
Abdominal pain 67 6.67
Figure 2: Visualize information about selected health centres
in LGAs
3. DESIGN OF EXPERIMENT AND
2.2 Data Analysis & Interpretation RESULTS
Statistically, out of the 1,006 patient’s data captured, 44.8% were 3.1 Research Experimental Platform
male while the remaining 55.2% are female, (See Table 2). The Waikato Environment for Knowledge Analysis (WEKA) platform
analysis further revealed that 57.1% of the patients are exposed to was used for the data analytic experiment. It is a powerful data
chemical and industrial contents while 32.7% of the population mining tool that has a GUI Chooser from which any of the four
as gender and hereditary disorder. The patient’s life style data major WEKA application environments (Explorer, Experimenter,
collected also indicated that the people around this region are KnowledgeFlow and Simple CLI) can be selected. The Explorer
addicted to smoking and drinking of alcohol, regular use of non- Application is selected for this experiment because it has a
steroidal anti-inflamatory drug (NSAIDs) such workbench that contains a collection of visualization tools, data
as ibuprofen and naproxen, which can double the risk of the processing, attribute ranking and predictive modeling with
disease by 51%. Other factors include obesity; faulty genes; a graphical user interface (GUI) for easy access to this
41
functionalities, which are very important to the research work. As shown in Figure 3, the patient’s databank component is
WEKA is a collection of machine learning algorithms for data responsible for the data collection, updating and storing patient’s
mining tasks. Algorithms implemented in WEKA include: data from different sources. The classifier system component is
Bayesian classifiers, Decision Trees, Rules, Artificial Neural responsible for the data modeling based on the algorithms in the
Network (Functions), Lazy classifiers and miscellaneous layers. The performance evaluation component is responsible for
classifiers. But for the purpose of this work Rule Induction and the evaluation of the performance of the algorithms considered in
Decision Tree classifiers was considered. These families of the layers using standard metric to produce the best (optimal)
classifiers have been selected because of their performances in algorithm. The rule generated from this algorithm is to be
various domains. They have both been successfully applied to a incorporated into the prediction system. Since the objective of the
variety of real-world classification tasks in industry, business, research work is to present a suitable algorithm for the cancer of
science and education with good performances [10]. The classifier the kidney prediction system, which the work has achieved. Hence
system designed for the data modeling as shown in Figure 3 is of the prediction system processes is not discussed in the work, but
two layers: Layer 1 consists of JRiP, PART and Decision Table of will be discussed in the future work of this research.
the family of Rules Induction and Layer 2 consists of J48, LAD
Tree, Decision Stump, Random Forest, Rep Tree, BF Tree, and 3.2 Experimental Results
LMT from the family of Decision Tree. The Decision Tree also Ten (10) classification algorithms from the family of classifiers
known as “white box” classification model can provide implemented in this work were used to model the patient’s
explanation for their models, and could be used directly for dataset. The datasets for the experiment was first divided into two,
decision making [5], while the Rule Induction is one of the which includes the training and testing datasets. 66% of the
fundamental tools of data mining, in which formal rules are datasets was devoted to training while the remaining 34% was
extracted from a set of observations. The rules extracted represent used for testing of randomly selected data. JRip, PART and
a full scientific model of the data [6]. According to Kapil et al., Decision Table in layer 1 of the classifier system were first used to
(2013), rule induction is a popular and well researched method for model the patient’s data and later the Decision Tree classifiers.
discovering interesting relations between variables in large The 10-fold cross validation test and percentage split modes were
database. These abilities and aptitudes of rule induction are suited also considered in the modeling. Since they are from different
and of good requirement for any effective and efficient intelligent classifiers family, they yielded different models that classify
system. A major paradigm of the Rule Induction is the differently on some inputs. The algorithms were tested on the
Association Rules [7]. datasets in order to determine that which best models the data
with best predictive accuracy.
Classifier System The comparison of the performance of the various algorithms in
layer 1 and layer 2 based on the output from the percentage split
Patient’s
Layer 1 Layer 2 (hold-out) and 10-fold cross validation modes was carried out.
Databank Rule Induction Decision Tree The results of the models from the two modes and the
performance evaluations are presented in Table 3. The 10-fold
cross-validation test mode was considered good since it produced
Performance the best model both in layer 1 and 2 of the classifier system.
Optimal
Evaluation Moreover, the 10-fold cross validation mode have been widely
Algorithm
used, and it is described a better option to determine the
performance of a classifier [8]. Table 4 shows the standard metric
accuracy details from the 10-fold cross validation mode
KC Prediction considered for all the algorithms in the experiment. Figure 4 and
System Figure 5 show the graphs of predictive accuracy and time taken to
build the models by the classifiers respectively.
Figure 3: Designed Classifier System
42
Table 3: Classification Accuracy Comparison between Hold-out and 10-fold Cross Validations in Layer 1 and Layer 2
10-fold Cross Validation Hold-out (Percentage Split)
Correctly Classified Time taken to Correctly Classified Time taken to
S/N Classifiers Instances build model Instances build model
1 J48 Decision Tree 74.7 0.03 74.5 0.02
2 LMT 74.6 29.25 73.7 29.03
3 LAD Tree 72.6 0.92 73.1 0.91
4 RepTree 71.6 2.54 72.4 2.4
5 JRiP Rules 70.9 0.03 70.1 0.03
6 PART 70.8 0.02 71.8 0.03
7 Decision Table 70.2 0.03 70.3 0.03
8 Random Forest 69.6 0.13 70.7 0.11
9 Decision Stump 64.7 0.01 64.9 0.01
10 BF Tree 57.9 2.54 60.8 2.55
Table 4: Compared standard metric accuracy details for all the Classification Algorithms
S/N Algorithms TP FP Precision Recall F- ROC Built Correctly
Rate Rate Measure Area Time(s) classified %
1 J48 Decision Tree 0.747 0.135 0.687 0.714 0.614 0.78 0.03 74.7
2 LMT 0.746 0.239 0.73 0.746 0.733 0.863 29.25 74.6
3 LAD Tree 0.731 0.292 0.714 0.731 0.702 0.85 0.91 73.1
4 RepTree 0.716 0.548 0.536 0.658 0.533 0.571 0.03 71.6
5 JRiP 0.709 0.274 0.728 0.749 0.731 0.754 0.06 70.9
6 PART 0.718 0.294 0.694 0.718 0.695 0.814 0.03 71.8
7 Decision Table 0.704 0.238 0.716 0.704 0.702 0.816 0.05 70.4
8 Decision Stump 0.649 0.36 0.579 0.647 0.612 0.669 0.02 64.9
9 Random Forest 0.643 0.327 0.622 0.643 0.629 0.74 0.08 64.3
10 BF Tree 0.579 0.223 0.718 0.716 0.717 0.748 2.54 57.9
Figure 5: Time Taken by the Classifiers to build Models in
Figure 4 Predictive Accuracy of Classifiers in Layers 1 and 2 Layers 1 and 2 for both 10-fold cross validations and
for both 10-fold and Hold-out (Percentage Split) Validations percentage Split (hold- out)
43
From the experimental results and analysis, it shows that the J48 Rule 3: IF (G&H Disorder = NO) AND (C&I Exposure = Yes)
decision tree and LMT rules outperform all other algorithms in AND (Lifestyle = Smoking) AND Complaints = tumor: PL =
the layers. However, J48 decision tree was chosen as the best Three
algorithm in this work because it has the correctly classified
instances of 74.7%, ROC Area of 0.78 and recall of 0.714
respectively. It has a lower FP rate of 0.153, F-Measure of 0.614 Rule 4: IF (G&H Disorder = NO) AND (C&I Exposure = Yes)
and took lesser time of 0.03 seconds to build the model compared AND (Lifestyle = Smoking) AND Complaints = Fibroids: PL =
to LMT and other classifiers as shown in Table 4. Additionally, Three
J48 decision tree algorithms generally have this ability that can Rule 5: IF (G&H Disorder = NO) AND (C&I Exposure = Yes)
produce a simple tree structure with high accuracy in term of AND (Lifestyle = Smoking) AND Complaints = Stomach ucher :
classification rate, even with huge volume of data [9]. Pruning PL = Two
methods have been introduced to reduce the complexity of tree
structure without any decrease in classification accuracy. The J48 Rule 6: IF (G&H Disorder = NO) AND (C&I Exposure = Yes)
decision tree structure and rules as generated by WEKA are AND (Lifestyle = Smoking) AND Complaints = Kidney pain:
presented in Figure 6. One
Rule 7 IF (G&H Disorder = NO) AND (C&I Exposure = Yes)
AND (Lifestyle = Smoking) AND Complaints = Abdominal pain:
Two
Rule 8 IF (G&H Disorder = YES) AND (C&I Exposure = Yes)
AND (Lifestyle = Smoking) AND Complaints = blood in urine:
PL = One
Rule 9 IF (G&H Disorder = YES) AND (C&I Exposure = Yes)
AND (Lifestyle = Obesity) AND Complaints = blood in urine: PL
= Two
Rule 10 IF (G&H Disorder = YES) AND (C&I Exposure = Yes)
AND (Lifestyle = HB Pressure) AND Complaints = blood in
urine: PL = Two
Rule 11 IF (G&H Disorder = YES) AND (C&I Exposure = Yes)
AND (Lifestyle = Smoking) AND Complaints = Drug Abuse OR
Tumor OR Fibroids: PL = Two
Rule 12 IF (G&H Disorder = YES) AND (C&I Exposure = Yes)
AND (Lifestyle = Smoking) AND Complaints = Abdominal pain:
PL = Two
Rule 13 IF (G&H Disorder = YES) AND (C&I Exposure = Yes)
Figure 6: J48 Decision Tree Structure as presented by WEKA AND (Lifestyle = Smoking) AND Complaints = Kidney pain: PL
= One
The rules generated from the best algorithm (J48 pruned decision
tree) are as stated in rules 1 to 20. The rules were tested in a Rule 14 IF (G&H Disorder = YES) AND (C&I Exposure = Yes)
prediction system framework and their prediction levels are AND (Lifestyle = Smoking) AND Complaints = stomach ucher:
classified as follows: (PL) – One, Two and Three. This show the PL = One
status of patients and by interpretation: Level One and Two Rule 15 IF (G&H Disorder = YES) AND (C&I Exposure = Yes)
indicates a risk level or status of the disease manifestation in the AND (Lifestyle = Alcohol OR Dialysis) AND Complaints =
patients that needs to be attended to urgently. While, level Three stomach ucher: PL = Two
indicates that the patient is not manifesting any symptoms of
Rule 16 IF (G&H Disorder = YES) AND (C&I Exposure = Yes)
kidney cancer disease, but may suffer from other diseases. A
AND (Lifestyle = Radiation) AND Complaints = stomach ucher
back-end for updating the rules as the situation arises will be
OR blood in urine: PL = One
incorporated into the system to match other conditions.
Rule 17 IF (G&H Disorder = YES) AND (C&I Exposure = Yes)
AND (Lifestyle = Water pills) AND Complaints = stomach ucher:
Rule 1: IF (G&H Disorder = NO) AND (C&I Exposure = Yes) PL = Three Rule 18 IF (G&H Disorder = YES) AND (C&I
AND (Lifestyle = Smoking) AND Complaints = blood in urine: Exposure = NO) AND (Lifestyle = Smoking) AND Complaints =
PL = One stomach ucher OR kidney pain: PL = One
Rule 2: IF (G&H Disorder = NO) AND (C&I Exposure = Yes)
AND (Lifestyle = Smoking) AND Complaints = back pain: PL = Rule 19 IF (G&H Disorder = YES) AND (C&I Exposure = NO)
Two AND (Lifestyle = Smoking) AND Complaints = stomach ucher:
PL = Two
44
Rule 20 IF (G&H Disorder = YES) AND (C&I Exposure = NO) 5. REFERENCES
AND (Lifestyle = Smoking OR Obesity OR Drug Abuse OR
Radiation OR Water Pills OR Dialysis) AND Complaints = [1] Lasebikan OA, Nwadinigwe CU, Onyegbule EC Pattern of
stomach ucher: PL = Three bone tumours seen in a regional orthopaedic hospital in Nigeria.
[2] Kushi LH, Doyle C, McCullough M, et al. (2012). "American
4. CONCLUSIONS Cancer Society Guidelines on nutrition and physical activity for
The research work was focused at presenting an efficient cancer prevention: reducing
algorithm suitable for predicting the status of kidney cancer in
[3] Kohavi, R., Rothleder, N. J’, & Simoudis, A.P (2002):
patients. To achieve the objectives of the research work: (i).
Emerging Trends in Business Analytics Published by ACM
Dataset pertaining to patient was acquired from fifty LGA (52)
Volume 45 Issue 8, Pages 45-48 August 2002.
selected Health Centres in the south western region of Nigeria
using purposive and selective sampling techniques. (ii) the [4] Berson, Smith ad Thearling ((199)
researcher developed a two-layered classifier system consists of [5] Romero, C., Olmo, J. L & Ventura, S (2013): A meta-learning
Rule Induction and Decision Trees implemented on Waikato approach for recommending a subset of white-box classification
Environment for Knowledge Analysis (WEKA) to build the data algorithms for Moodle datasets. Department of Computer
model using data analytic approach, and (iii) different machine Science, University of Cordoba, Spain.
learning algorithms were used in search for the algorithm that
produced the best model with predictive accuracy. In the [6] Grzymala-Busse, J. W (2013). Rule Induction - University of
experiment, ten (10) classification model algorithms from Kansas. Extracted 20-06-2013.
different classifier family were implemented on the [7] Kapil, S., Sheveta, V., Heena, S., Richa, D & Jasreena, K. B
patients’dataset. Since they are from different classifiers family, (2013). A Hybrid Approach Based On Association Rule Mining
they yielded different models that classify differently on some and Rule Induction in Data Mining International Journal of Soft
inputs. The comparison of the performance of the various Computing and Engineering (IJSCE) ISSN: 2231-2307, Volume-
algorithms in layer 1 and layer 2, and the standard metrics of 3, Issue-1, March 2013 146.
accuracy, precision, recall and f-measure for the best classifier [8] WEKA,(2011): WEKA Tutorial. The University of Waikato
considered in this work was carried out as shown in Table 3 and (2011). Available at: http://www.cs.waikato.ac.nz/ml/weka/,
Table 4 respectively. The results show that the J48 decision tree (Accessed 20 July, 2013).
outperform all other algorithms in the layers with predictive
accuracy of correctly classified instances of 74.7 % in 0.03 [9] Mohamed, W. Nor Haizan W, Mohd N. S, & Abdul H. O
seconds, ROC Area of 0.78, FP rate of 0.153, TP rate of 0.714, (2012). A Comparative Study of Reduced Error Pruning Method
precision and recall of 0.614. in Decision Tree Algorithms. IEEE International Conference on
Control System, Computing and Engineering, 23 - 25 Nov. 2012,
Penang, Malaysia
.
45