weka data mining
- 1. K A LT H O O M A L M A Q B A L I & M AT T C O U RT N E Y WEKA WEB AND SOCIAL COMPUTING
- 2. OUTLINES • Weka . • REPTree results and compare it with different trees. • REPTree analysis. • JRIP results and compare it with different rules . • JRIP rule analysis. • Conclusion . • Question & Answers .
- 3. WEKA A collection of machine learning algorithm for data mining tasks. It is also well-suited for developing new machine learning schemes. The algorithms can either be applied directly to a dataset or called from your own Java code.
- 4. WEKA Weka contains tools for : • Data pre-processing: called “filters” ,for discretization, normalization, resampling, attribute selection, transformation and combination of attributes. • Data classification: for predicting nominal or numeric quantities . • Data clustering : for finding groups of similar instances in a dataset. • Data association : an implementation of the Apriori algorithm for learning association rules. • Attribute Selection :searches through all possible combinations of attributes in the data and finds which subset of attributes works best for prediction • Data visualization :allows you to visualize a 2-D plot of the current working relation , it is useful in practice, which helps to determine difficulty of the learning problem.
- 5. REDUCED ERROR PRUNING TREE (REPTREE) • Uses regression tree logic. • Splits data based on information gain or variance reduction. • Creates multiple trees, then selects best. • Prunes selected best with reduced error pruning. • Results in a “fast” learning algorithm.
- 6. TEST DATA
- 8. GRAPH - MEAN CORRECT CLASSIFICATIONS (%)
- 9. GRAPH - RANGE CORRECT CLASSIFICATIONS (%)
- 10. GRAPH - MEAN TIME (MS)
- 11. AGGREGATE DATA - RANKED
- 12. AGGREGATE DATA - RANKED
- 13. WHAT DOES THIS MEAN?
- 14. WHAT DOES THIS MEAN? • REPTree is a fast learning algorithm. • REPTree is 5x faster than J48 (average). • J48 results in 1.05x more correct classifications (average). • May be useful for initial passes of datasets to look for potential knowledge gain with good potential for success. More likely than J48 to do this in excellent time.
- 15. JRIP RULE • Is a rule-based learner that builds a set of rules that identify the classes while Repeated Incremental Pruning to Produce Error Reduction. • This algorithm was designed by Cohen in 1995 namely • JRIP is especially more efficient on large noisy datasets . • There are two kinds of loop in JRip algorithm: outer loop which adds one rule at a time to the rule base and Inner loop adds one condition at a time to the current rule.
- 16. •
- 17. TEST DATA •
- 18. AGGREGATE DATA
- 19. AVERAGE GRAPH •
- 20. • What I found out after applied JRIP and compared it with other rules results ?
- 21. WHAT I FOUND ? • JRIP rule is a fast learning algorithm. • JRIP got good correct range in both mean and range time )compare to other rules). • JRIP is not the highest , not the lowest ( average ). • Do I recommend to use JRIP rule ?
- 22. Q& A