Code review automation using machine learning

V tem magistrskem delu smo se posvetili področju pregledovanja kode s pomočjo strojnega učenja. Proučili smo sorodna dela na tem področju ter določili teoretični pristop, s pomočjo katerega bomo lahko izvajali napovedovanje slabih sprememb programske kode programskega jezika Javascript, ki zahtevajo odpravo napak. Tako bomo zmanjšali porabo virov pri pregledovanju programske kode. Izdelali smo prototip ekspertnega sistema, ki bo omogočal generiranje metrik in učenje nevronske mreže v ogrodju Tensorflow.js. Učinkovitost sistema smo ovrednotili na treh odprtokodnih projektih ter dosegli rezultate, ki upravičujejo smiselnost vpeljave takšnega sistema v proces razvoja programske opreme.This master thesis adresses code review process supported by machine learning. We studied works of other authors for suitable indicators to generate theoretical approach, which can be used to predict rework in Javascript programming language. This way we can assist code review process by using less resources. We developed an expert system prototype which generates needed metrics to perfrom machine learning using Tensorflow.js library. Developed system was validated for effectiveness on three opensource projects, which proved usefullness of predictions and helpful contribution to software development process