Subgroup Discovery for Defect Prediction

Subg oup Disco e y o De ec P edic ion Daniel Rod ´ıguez1,R.Ruiz 2, J.C. Riquelme3, and Rachel Ha ison4 1Uni . o Alcal´a, 28871 Alcal´a, Spain [email p o ec ed] 2Pablo de Ola ide Uni ., 41013 Se ille, Spain [email p o ec ed] 3Uni . o Se ille, 41012 Se ille, Spain [email p o ec ed] 4Ox o d B ookes Uni ., Ox o d OX33 1HX, UK [email p o ec ed] Al hough he e is ex ensi e li e a u e in so wa e de ec p edic ion echniques, machine lea ning app oaches ha e ye o be ully explo ed and in pa icula , Subg oup Disco e y (SD) echniques. SD algo i hms aim o find subg oups o da a ha a e s a is ically diffe en gi en a p ope y o in e es [1,2]. SD lies be ween p edic i e (finding ules gi en his o ical da a and a p ope y o in e es ) and desc ip i e asks (disco e ing in e es ing pa e ns in da a). An impo an diffe ence wi h classifica ion asks is ha he SD algo i hms only ocus on finding subg oups (e.g., inducing ules) o he p ope y o in e es and do no necessa ily desc ibe all ins ances in he da ase . In his p elimina y s udy, we ha e compa ed wo well-known algo i hms, he Subg oup Disco e y algo i hm [3] and CN2-SD algo i hm [4], by applying hem o se e al da ase s om he publicly a ailable PROMISE eposi o y [5], as well as he Bug P edic ion Da ase c ea ed by D’Amb os e al. [6]. The compa ison is pe o med using quali y measu es adap ed om classifica ion measu es. The esul s show ha gene a ed models can be used o guide es ing effo . The pa ame e s o he SD algo i hms can be adjus ed o balance he specifici y and gene ali y o a ule so ha he selec ed ules can be conside ed good enough o so wa e enginee ing s anda ds. The induced ules a e simple o use and easy o unde s and. Fu he wo k wi h mo e da ase s and o he SD algo i hms ha ackle he disco e y o subg oups using diffe en app oaches (e.g., con inuous a ibu es, disc e iza ion, quali y measu es, e c.) is needed. Re e ences 1. W obel, S.: An algo i hm o mul i- ela ional disco e y o subg oups. In: P oceedings o he 1s Eu opean Symposium on P inciples o Da a Mining, pp. 78–87 (1997) 2. He e a, F., Ca mona del Jesus, C.J., Gonz´alez, P., del Jesus, M.J.: An o e iew on subg oup disco e y: Founda ions and applica ions. Knowl. In . Sys . (2010) Resea ch suppo ed by Spanish Minis y o Educa ion MEC TIN 2007-68084-C02. 3. Gambe ge , D., La ac, N.: Expe -guided subg oup disco e y: me hodology and applica ion. Jou nal o A ificial In elligence Resea ch 17, 501–527 (2002) 4. La aˇc, N., Ka ˇsek, B., Flach, P., Todo o ski, L.: Subg oup disco e y wi h CN2-SD. The Jou nal o Machine Lea ning Resea ch 5, 153–188 (2004) 5. Boe iche , G., Menzies, T., Os and, T.: P omise eposi o y o empi ical so wa e enginee ing da a. Wes Vi ginia Uni e si y, Depa men o Compu e Science (2007) 6. D’Amb os, M., Lanza, M., Robbes, R.: An ex ensi e compa ison o bug p edic ion app oaches. In: IEEE Mining So wa e Reposi o ies (MSR), pp. 31–41 (2010)