Univerza na Primorskem Fakulteta za matematiko, naravoslovje in informacijske tehnologije

četrtek, 29. avgust 2019 How overall coverage of an association rule learning classifier affects its accuracy?

V ponedeljek, 2. septembra 2019, bo ob 16. uri v prostorih Fakultete za matematiko, naravoslovje in informacijske tehnologije Univerze na Primorskem (Glagoljaška 8, Koper)
Oddelkov za Informacijske znanosti in tehnologije UP FAMNIT in UP IAM.

ČAS/PROSTOR: 2. september 2019 ob 16.00 v FAMNIT-VP2


Jamolbek Mattiev has a master degree in computer systems and software from the National  University of Uzbekistan. He is currently finishing the second year of the PhD program Computer Science at UP FAMNIT and is employed as teaching assistant on the BSc Computer Science program at UP FAMNIT. He is doing his research in the fields of Data Mining and Machine Learning.

NASLOV: How overall coverage of an association rule learning classifier affects its accuracy?

Classification and association rule are two important technologies in data mining. Researchers have recently proposed several classification techniques based on the concept of association rules (also known as CBA-based methods). Experimental results show that in average the CBA-based approaches could achieve higher accuracy than some of traditional classification methods. Associative classification (AC) is a data mining approach that combines classification and association rule to build classification models (classifiers).
In this seminar, we will present associative classification, where class association rules are generated and analyzed to build a simple, compact, understandable and relatively accurate classifier. Furthermore, we find the overall coverage and average rule coverage of the classifier that affects its the accuracy. We compare the accuracies our method that uses constrained exhaustive search with that of some “classical” classification rule learning algorithms that use greedy heuristic search on some “real-life” datasets. We have performed experiments on 11 datasets from UCI Machine Learning Database Repository. Overall coverage shows the power of the classifier. Experimental evaluations show that our proposed method outperforms Naive Bayes and C4.5 on average accuracy.


Predavanje bo v angleškem jeziku.