Jagdeep Singh HYBRID TECHNIQUE FOR ASSOCIATIVE CLASSIFICATION OF HEART DISEASES
Jan 19, 2015
Jagdeep Singh
HYBRID TECHNIQUE FOR ASSOCIATIVE CLASSIFICATION
OF HEART DISEASES
Table of Contents
Ø Introduction Ø Motivation Ø Data Mining Ø Classification Ø Association Ø Heart Disease Database
Ø Literature Survey Ø Problem Formulation Ø Objectives
Ø Present Work Ø Result and Discussion Ø Conclusion Ø Future Scope Ø References
Motivation
Ø Accumulation of huge data-sets in the field of Engineering and Biomedical Science.
Ø Ability to extract hidden and useful knowledge from large databases.
Ø Need to development intelligent and cost effective decision support system.
Ø How to teach the people to ignore the irrelevant data.
Ø The greatest problem of today is to get optimal outcome of irrelevant data.
Data Mining
Ø Data mining computational process of finding patterns in large data sets including methods at the intersection of machine learning, artificial intelligence, statistics and database systems.
Ø The main focus of data mining process is to obtain information from the data and converted it into an knowledgeable and reasonable structure for further use.
Data Mining Process
The Data Mining Process [1]
Classification
Classification is the problem of identifying to which of a set of categories a new observation belongs, on the basis of a training set of data containing observations (or instances) whose category membership is known.
Association
Association learning method for discovering interesting relations between variables in large databases. It is intended to identify strong rules discovered in databases using different measures of interestingness.
For example, the rule : {onions, potatoes} => {burger}.
Example : Heart diseases Dataset
ID age Gender Chest pain Blood pressure diagnosis
1 63 male typ_angina High No
2 67 male asympt very_high Yes
3 67 male asympt high Yes
4 37 male non_anginal high No
5 41 female atyp_angina high No
6 56 male atyp_angina high No
7 62 female asympt high Yes
8 57 female asympt high No
9 63 male asympt high Yes
10 53 male asympt high Yes
11 57 male asympt high No
12 56 female atyp_angina high No
13 56 male non_anginal high Yes
14 44 male atyp_angina high No
Association rules example:
1. cp=atyp_angina trestbps=high 4 ==> diagnosis=No 4
2. gender=male cp=asympt trestbps=very_high 2 ==> diagnosis=Yes 1
3. gender=female cp=atyp_angina 2 ==> diagnosis=No 2
4. gender=male cp=atyp_angina trestbps=high 2 ==> diagnosis=No 2
5. gender=female cp=atyp_angina trestbps=high 2 ==> diagnosis=No 2
6. cp=atyp_angina 4 ==> diagnosis=No 4
7. gender=male cp=asympt trestbps=high 4 ==> diagnosis=Yes 2
8. gender=male cp=atyp_angina 2 ==> diagnosis=No 2
Result new prediction ?
age gender Chest pain
Blood pressure
diagnosis
52 male non_anginal very_high
Classifiers
Ø ZeroR : There is no predictability, it is useful for determining a baseline performance as a benchmark for other classification methods.
Ø OneR : Classification rules based on the value of a single predictor, that generates one rule for each predictor in the data.
Ø NaiveBayes: Bayes rule is implemented or assigned to make easier to evaluate prior from a probability model. it handles condition of some missing entries in data.
Ø J48: It creates a binary tree, With this technique, a tree is constructed to model the classification process.
Ø IBk (k nearest neighbour): The nearest neighbor algorithm categorise a given instance depend on a set of already categorise the training set by measuring the distance to the closed instances
Association Methods
Ø Aprior Algorithm: Find rules that will predict the occurrence of an item based on the occurrences of other items in the transaction.
Ø FP-Growth Algorithm: Allows frequent discovery
without candidate itemset generation. Extracts frequent itemsets form the FP-tree. Follow Divide and conquer approach.
Heart Disease Database
Sr. No.
Attributes
Description
Values
1 age Age in years Continuous
2 gender Male or female 1 = Male, 0 = female
3 cp Chest pain type
1 = typical type, 2 = typical type angina, 3 = non-angina pain, 4 = asymptomatic
4 thestbps Resting blood pres- sure Continuous value in mm hg
5 chol Serum cholesterol Continuous value in mm/dl
6 thalach Maximum heart rate achieved Continuous value
7 fbs Fasting blood sugar 1 =>120 mg/dl, 0 =<120 mg/dl
Continue…
8 Restecg Resting electro- graphic results
0 = normal, 1 = having ST-T wave abnormal, 2 = left ventricular hypertrophy
9 exang Exercise induced angina 0 = no 1 = yes
10 oldpeak ST depression induced by exercise relative to rest
Continuous value
11 slope Slope of the peak exercise ST segment
1 = unsloping, 2 = flat, 3 = downsloping
12 ca Number of major vessels colored by floursopy
0 - 3 value
13 thal Defect type 3 = normal, 6 = fixed, 7 = reversible defect
14 Diagnosis Heart disease Predi- cation
Value 1: no heart disease Value 0: has heart disease
Literature Survey
Ø Liao et al. [3] author report about data mining techniques and application,
development through a survey of literature, form 2000 to 2011. Paper surveys
three areas of data mining research: knowledge types, analysis types, and
architecture types. A discussion deals with future progress in social science and
Engineering methodologies implement data mining techniques and the development
of applications in problem- oriented
Ø Liu et al. [4] presented an associative classification, to integrate classification rules
and association rule mining. The integration is done by focusing on mining a special
subset of association rules whose consequent parts are restricted to the classification
class labels, called Class Association Rules (CARs). This algorithm first generates all
the association rules and then selects a small set of rules to form the classifiers.
When predicting the class label for a coming sample, the best rule is chosen.
Continue…
Ø The first association rule mining algorithm was the Apriori algorithm [5] developed
by Agrawal, and swami. The Apriori algorithm generates the candidate item sets in
one pass through only the item sets with large support in the previous pass, without
considering the transactions in the database.
Ø Palaniappan and Awang [6] developed a prototype Intelligent Heart Disease
Prediction System (IHDPS) using data mining techniques, namely, Decision Trees,
Nave Bayes and Neural Network. Results show that each technique has its unique
strength in realizing the objectives of the defined mining goals. IHDPS can answer
complex what if queries which traditional decision support systems cannot. Using
medical profiles such as age, gender, blood pressure and blood sugar it can predict
the likelihood of patients getting a heart disease. IHDPS is Web-based, user-
friendly, scalable, reliable and expandable. It is implemented on the .NET platform.
Continue…
Ø Srinivas et al. [7] presented Application of Data Mining Technique in Healthcare and Prediction of Heart Attacks. The potential use of classification based data mining techniques such as Rule based, Decision tree, Nave Bayes and Artificial Neural Network to the massive Volume of healthcare data. Tanagra data mining tool was used for exploratory data analysis, machine learning and statistical learning algorithms. The training data set consists of 3000 instances with14 different attributes.
Ø Shouman et al. [8] proposed k-means clustering with the decision tree method to predict the heart disease. In their work they suggested several centroid selection methods for k- means clustering to increase efficiency. The 13 input attributes were collected from Cleveland Clinic Foundation Heart disease data set. For the random attribute and random row methods, ten runs were executed and the average and best for each method were calculated. In Addition, integrating k-means clustering and decision tree could achieve higher accuracy than the paging algorithm in the diagnosis of heart disease patients. The accuracy achieved was 83.9% by the enabler method with two clusters.
The algorithm used Accuracy Time taken Naive Bayes 52.33% 609ms Decision list 52% 719ms K-NN 45.67% 1000ms
Summary and Gaps Identified
Ø Implementation of different methods like NaiveBayes, Decision tree and Neural, K-nearest, Artificial Neural Network etc, is done on heart disease dataset.
Ø The performance of the classifiers is evaluated and their results are analysed.
Ø Maximum accuracy achieved according to the survey is 83.9% using K-means clustering with decision tree.
Ø The classification methods does not provide better accuracy and
experimental results.
Ø Integration of associative classification is not yet implemented on heart diseases data set.
Problem Formulation
Ø Accuracy of heart data diseases is only calculate on basis of classification
methods.
Ø Accuracy of corrected classified instances is less to predict heart diseases.
Ø Association and classification suffers from inefficiency due to the fact that it
often generates a very large number of insignificant rules.
Ø Most of the associative classification algorithms adopt the exhaustive search
method to discover the rules and require multiple passes over the
database.
Ø They find frequent items in one phase and generate the rules in a separate
phase consuming more resources such as storage and processing time.
Objectives
Ø To propose a technique that can generate Classification Association Rules (CARs) efficiently for heart diseases prediction.
Ø Perform evaluation of proposed approach. Ø Comparative analysis of proposed method with
other state-of-the-art techniques
Present Work
The Present Work has been implemented using data mining tool Weka . Implementation steps are listed below :
1. Review of the classification and association rule generation methods. 2. Understanding the existing algorithm of classification. 3. Study the existing methods of Classification and association to predict heart diseases. 4. Understanding the heart disease data set attributes used in predication. 5. Study ARFF file format standard of representing datasets. 6. Preparing data set for implementation of association algorithm
Continue…
7. Implement association algorithm like Aprior and FP growth on prepared data set. 8. Select the best 10 rules for each associate algorithm. 9. Make classes and extract training data sets bases on different rules. 10. Implement classification algorithms on extracted training data set. 11. Compared the performance and accuracy of corrected classified instances of classification methods. 12. Construct a system based on high performance and better accuracy of classification meth- ods.
Apriori algorithm best rules
1. gender=female fbs=f restecg=normal exang=no thal=normal 35 ==>diagnosis=No 35 conf:(1).
2. gender=female cp=non anginal thal=normal 31 ==>diagnosis=No 31 conf:(1).
3. cp=asympt chol=high risk thal=reversable defect 42 ==>diagnosis=Yes 41 conf:(0.98)
4. cp=asympt restecg=left vent hyper thal=reversable defect 41 ==>diagnosis=Yes 40 conf:(0.98)
5. gender=female fbs=f slope=up 39 ==>diagnosis=No 38 conf:(0.97)
6. gender=female restecg=normal exang=no thal=normal 38 ==>diagnosis=No 37 conf:(0.97)
7. gender=female fbs=f restecg=normal exang=no 37 ==>diagnosis=No 36 conf:(0.97)
8. gender=female fbs=f slope=up thal=normal 37 ==>diagnosis=No 36 conf:(0.97)
9. cp=asympt trestbps=high chol=high risk thal=reversable defect 37 ==>diagnosis=Yes 36 conf: (0.97). 10. gender=female cp=non anginal 35 ==>diagnosis=No 34 conf:(0.97).
FP-Growth algorithm best rules 1. (fbs binarized=1, restecg=left vent hyper binarized=1, diagnosis=Yes, exang binarized =1): 31 ==>(cp=asympt binarized=1): 31 conf:(1)
2. (chol=high risk binarized=1, cp=asympt binarized=1, thal= reversable defect binarized = 1): 42 =>(diagnosis=Yes): 41 conf:(0.98)
3. (restecg=left vent hyper binarized=1, cp=asympt binarized=1, thal= reversible defect bi- narized =1): 41 ==>(diagnosis=Yes): 40 conf:(0.98)
4. (thal=normal binarized=1, trestbps=normal binarized=1): 37 ==>(fbs binarized=1): 36 conf:(0.97)
5. (slope=up binarized=1, thal=reversable defect binarized=1): 37 ==>(gender binarized=1): 36 conf:(0.97)
6. (trestbps=high binarized=1, chol=high risk binarized=1, cp=asympt binarized=1, thal= re- versable defect binarized=1): 37 ==>(diagnosis=Yes): 36 conf:(0.97)
7. (chol=high risk binarized=1, thal=reversable defect binarized=1, exang binarized=1): 34 ==>(diagnosis=Yes): 33 conf:(0.97)
8. (fbs binarized=1, chol=high risk binarized=1, cp=asympt binarized=1, thal= reversible defect binarized=1): 34 ==>(diagnosis=Yes): 33 conf:(0.97)
9. (gender binarized=1, chol=high risk binarized=1, cp=asympt binarized=1, thal= reversible defect binarized=1): 34 ==>(diagnosis=Yes): 33 conf:(0.97)
10. (fbs binarized=1, restecg=left vent hyper binarized=1, cp=asympt binarized =1, thal= re- versable defect binarized=1): 33 ==>(diagnosis=Yes): 32 conf:(0.97)
Sample Data form of Heart Disease Prediction Online Available : http://gndec.ac.in/~jagdeepmalhi/ihdps/
Sample Data of Heart Disease Prediction for Risk Level: No
Sample Data of Heart Disease Prediction for Risk Level: Low
Sample Data of Heart Disease Prediction for Risk Level: High
Results and Discussion
The Evaluation of results is done on bases of two categories.
Ø Compare the different parameters like time taken, Correctly/Incorrectly classified instances, Kappa statistic value, mean absolute error and root mean squared error rate of different classifier with Aprior and FP-Growth association algorithm.
Ø Compare the accuracy evaluated by different authors on the heart disease dataset.
Continue…
Comparison of different classifiers using Aprior association algorithm on heart diseases dataset.
Classifiers Time Taken (In seconds)
Correctly C l a s s i f i e d I n s t a n c e s (%)
Incorrectly C l a s s i f i e d I n s t a n c e s (%)
Kappa statistic
Mean absolute error
Root mean squared error
ZeroR 0.001 67.2 32.79 0 0.441 0.470
OneR 0.01 97.31 2.6 0.94 0.027 0.164
J48 0.04 97.85 2.15 0.951 0.031 0.143
IBk 0.003 99.19 0.81 0.982 0.010 0.090
NaiveBayes 0.01 97.58 2.42 0.946 0.023 0.137
Continue…
Comparison of different classifiers using FP- Growth association algorithm on heart diseases dataset.
Classifiers Time Taken (In seconds)
Correctly Classified Instances (%)
Incorrectly Classified Instances (%)
Kappa statistic
Mean absolute error
Root mean squared error
ZeroR 0.001 85.67 14.33 0 0.247 0.350
OneR 0.005 92.55 7.45 0.649 0.075 0.273
J48 0.01 96.56 3.44 0.859 0.056 0.185
IBk 0.001 94.84 5.16 0.779 0.053 0.227
NaiveBayes 0.003 97.55 7.45 0.711 0.088 0.265
Continue…
Comparison of Aprior and FP-Growth association algorithms heart diseases dataset
Association Algorithms
ZeroR accuracy
OneR accuracy
J48 accuracy
IBk accuracy
NaiveBayes accuracy
Aprior 67.2 97.31 97.85 99.19 97.58
FP-Growth 85.67 92.55 96.56 94.84 97.55
Continue…
Comparison of results evaluated by different authors on the heart disease dataset.
Author /Year Technique Accuracy (%)
Cheung 2001 [11] NaiveBayes 81.48
Polat and Sahan et al. 2007 [12] K-Nearest Neighbor 87.00
Shouman and Turner et al. 2012 [13] Decision tree 84.10
Das and Turkoglu et al. 2009 [14] K-Nearest Neighbor 97.40
Tu and Shin et al. 2009 [15] J4.8 Decision Tree 78.90
Proposed Method 2014 IBk with Aprior Algorithm 99.19
Conclusion
Ø The development of a hybrid technique for implementation of associative classification is done on heart diseases dataset to predict more accurate results.
Ø Dataset is implement on weka environment and compared the performance of different classifier after apply association algorithm.
Ø Results show that IBk (k Nearest Neighbor) with Aprior associative algorithms shows better results than others.
Ø Compare the results of different classifiers with proposed implementation methods.
Ø Finally develop Intelligent Heart Diseases Prediction System (IHDPS) for end user to check the risk of heart diseases.
Future Scope
Ø In future work plan to reduce numbers of attributes and to determine the attribute which contribute towards the diagnosis of heart disease.
Ø Additional Data Mining techniques can be incorporated to provide better results.
Ø There is a need to build a system where every human can check the risk of heart diseases using minimum recourses and parameters.
Ø Parameters like processing time, resources and memory used can be further enhanced.
References
1) U. Fayyad, G. Piatetsky-Shapiro, and P. Smyth, “Data mining to knowledge discovery in databases,” American Association for Artificial Intelligence, vol. 17, no. 3, pp. 37–54, 1996.
2) D. Aha. (1988, July) Heart disease databases. [Online]. Available: http://repository.seasr. org/Datasets/UCI/arff/heart-c.arff.
3) S. H. Liao, P. H. Chu, and P. Y. Hsiao, “Data mining techniques and applications - a decade review from 2000 to 2011,” Elsevier Expert Systems with Applications, vol. 39, no. 1, pp. 11 303–11 311, 2012.
4) B. Liu, W. Hsu, and Y. Ma, “Integrating classification and association rule mining,” In Knowledge Discovery and Data Mining, New York, vol. 2, pp. 80–86, 1998.
5) R. Agrawal and R. Srikant, “Fast algorithms for mining association rules,” in VLDB, Santi- ago, Chile, September 1994, pp. 487–499.
6) S.Palaniappan and R.Awang, “Intelligent heart disease prediction system using data mining techniques,” in IEEE/ACS International Conference, Doha, 2008, pp. 108–115.
7) K. Srinivas, B. K. Rani, and D. A. Govrdhan, “Application of data mining techniques in healthcare and prediction of heart attacks,” International Journal on Computer Science and Engineering, vol. 2, no. 2, pp. 250–255, 2011.
Continue …
8) M. Shouman, T. Turner, and R. Stocker, “Integrating decision tree and k-means clustering with different initial centroid selection methods in the diagnosis of heart disease patients,” in Proceedings of the International Conference on Data Mining, 2012.
10) J. Singh, H. Singh, and A. Kamra, “Recent trends in data mining: A review,” in Proceeding of 3rd International Conference on Biomedical Engineering and Assistive Technologies, Chandigarh, India, 2014, pp. 138–144.
11) N.Cheung, “Machine learning techniques for medical analysis,” B.Sc. Thesis, School of Information Technology and Electrical Engineering, University of Queenland, 2001.
12) K. Polat, S. Sahan, and S. Gunes, “Automatic detection of heart disease using an artifi- cial immune recognition system (airs) with fuzzy resource allocation mechanism and k-nn (nearest neighbor) based weighting preprocessing,” Expert Systems with Applications, pp. 625–663, 2007.
13) M. Shouman, T. Turner, and R. Stocker, “Applying k-nearest neighbor in diagnosing heart disease patients,” International Journal of Information and Education Technology, vol. 2, no. 3, pp. 220–223, June 2012.
14) R. Das, I. Turkoglu, and A. Sengur, “Effective diagnosis of heart disease through neural networks ensembles,” Expert Systems with Applications, Elsevier, pp. 7675–7680, 2009.
15) M. C. Tu, D. Shin, and D. Shin, “Effective diagnosis of heart disease through bagging approach,” in Proceeding of 2nd International Conference on Biomedical Engineering and Informatics. Seoul, South Korea: IEEE, October 2009, pp. 1–4.
Jagdeep Singh http://jagdeepmalhi.blogspot.com