Implementation and verification of an algorithm for pattern similarity estimation

Poslovanje u modernom svijetu svakim danom zahtijeva obradu sve većih skupova podataka. Porast procesorske moći računala i razvoj algoritama strojnog učenja omogućuju računalima samostalnu obradu takvih skupova podataka. Ovaj rad opisuje primjenu algoritama strojnog učenja u domeni tvorničkog testiranja elektroničke komunikacijske opreme. Rezultate testova potrebno je analizirati kako bi se smanjio broj proizvoda koji je nepotrebno proglašen neispravnim. Problem se svodi na problem klasifikacije rezultata testova na ispravne i neispravne. Strojno učenje je implementirano kao komponenta alata FRAG BDA, kako bi se ubrzao proces analize rezultata testova. U radu su opisani korišteni algoritmi, i analizirani njihovi rezultati i korisnost krajnjim korisnicima alata. Zaključeno je da se za opisani problem najbolji rezultati postižu algoritmom nasumične šume zajedno s algoritmima Simhash i Mutual information.Business in the modern world requires working with bigger and bigger data sets. The rise in processing power of computers and the development of machine learning algorithms allow computers to process such data sets all by themselves. This paper describes the use of machine learning algorithms in the production testing of electronic communication equipment. The results of these tests need to be analyzed in order to minimize the number of products wrongly declared faulty. The task is to classify failed test as false positives or actual faults in the products. Machine learning algorithms where developed as part of the FRAG BDA tool in order to speed up the analysis of test results. This paper describes machine learning algorithms, and compares their results and usefulness to the users. It is concluded that the random forest classifier used together with Simhash and Mutual information algorithms shows the best results for the described problem