Comparison of machine learning methods for prediction and classification in marketing

Prvi cilj rada je izrada modela za predviđanje u marketingu pomoću neuronskih mreža i stabla odlučivanja. Drugi cilj je usporedbom i testiranjem različitih arhitektura metoda strojnog učenja dobiti točniju metodu strojnog učenja. Nadalje, svrha rada je istražiti mogućnosti primjene metoda strojnog učenja u poslovnom odlučivanju u području marketinga Podaci preuzeti iz baze Kaggle klasificirani su i obrađeni prema uspješnosti prodaje putem telefona u programu Statistica. Također, u istraživačkom dijelu opisa i rezultata istraživanja podaci su obrađeni kroz izradu modela neuronskih mreža s testirane dvije aktivacijske funkcije: sigmoidnom i tangens-hiperbolnom te kroz izradu modela stabla odlučivanja. Rezultati dobiveni analizom i usporedbom najtočnijeg modela neuronskih mreža i stabla odlučivanja pokazali su da je model neuronskih mreža testiran pomoću logističke aktivacijske funkcije dao najtočniji rezultat klasifikacije. Iako zbog niske stope točnosti model još nije spreman za primjenu u praksi, moguća su dodatna poboljšanja. Znanstveni doprinos ovoga rada je izrada modela za predviđanje i klasifikaciju u marketingu uporabom metoda strojnog učenja. Dobiveni rezultati rada mogu koristiti za daljnja istraživanja metoda strojnog učenja.The first goal of the paper is to design a prediction model in marketing using neural networks and decision trees. The second goal is to provide a more accurate method of machine learning by comparing and testing different methods of machine learning. Furthermore, the purpose of the paper is to explore the possibilities of applying machine learning methods in business decision making in the field of marketing. The data downloaded from the Kaggle database is classified and processed according to the sales performance over the phone in the Statistica program. Also, in the research part of the description and the results of the research data were processed through the development of models of neural networks with two tested activation functions: sigmoid and tangent-hyperbolic and trough the development of the decision tree model. The results obtained by analyzing and comparing the most accurate model of neural networks and decision trees have shown that the model of neural networks that was tested by logistic activation function gave the most accurate classification result. Although due to the low accuracy rate model is not yet ready for use in practice, further improvements are possible. The scientific contribution of this paper is the development of a model for predicting and classifying in marketing with the use of machine learning methods. The results obtained can be used for further research of machine learning methods