Pregled bibliografske jedinice broj: 882835
Can company credit worthiness be predicted? – A Neural Network Approach
Can company credit worthiness be predicted? – A Neural Network Approach // 5th International Scientific Symposium Economy of Eastern Croatia - Vision and Growth / Mašek Tonković, Anka (ur.).
Osijek, Hrvatska: Ekonomski fakultet Sveučilišta Josipa Jurja Strossmayera u Osijeku, 2016. str. 198-207 (predavanje, međunarodna recenzija, cjeloviti rad (in extenso), znanstveni)
CROSBI ID: 882835 Za ispravke kontaktirajte CROSBI podršku putem web obrasca
Naslov
Can company credit worthiness be predicted? – A Neural Network Approach
Autori
Has, Adela ; Zekić-Sušac, Marijana
Vrsta, podvrsta i kategorija rada
Radovi u zbornicima skupova, cjeloviti rad (in extenso), znanstveni
Izvornik
5th International Scientific Symposium Economy of Eastern Croatia - Vision and Growth
/ Mašek Tonković, Anka - : Ekonomski fakultet Sveučilišta Josipa Jurja Strossmayera u Osijeku, 2016, 198-207
Skup
5th International Scientific Symposium Economy of Eastern Croatia - Vision and Growth
Mjesto i datum
Osijek, Hrvatska, 02.06.2016. - 04.06.2016
Vrsta sudjelovanja
Predavanje
Vrsta recenzije
Međunarodna recenzija
Ključne riječi
artificial neural networks, company creditworthiness, modelling, sensitivity analysis
Sažetak
The purpose of the paper is to create a prediction model of company creditworthiness by neural network methodology. Company creditworthiness has been investigated in previous research mostly using standard statistical modelling techniques, such as multiple and logistic regression. Due to their advantages, neural network methods have recently shown their success in many problem domains for prediction, classification, and association purposes. In this research, the artificial neural network as one of the machine learning method is used to model creditworthiness of Croatian companies. The input space consisted of 29 variables containing companies’ financial coefficients and additional variables such as defense interval (in days), days of accounts receivables, days of accounts payables, the number of employees, and other. Fourthy neural network architectures were tested in order to find the model which produces the smallest error and the stability of results. The most successful model yields the average classification rate of 84.57% in a 10-fold subsampling procedure. Besides the model accuracy, the paper also analyses the importance of predictors using sensitivity analysis. The results of suggested model are then compared to some previous research in this area and similar models in other countries. The research could be beneficial to business managers, investors, banks, government institutions, and other organizations that need information about company’s creditworthiness as an input for their decision making process.
Izvorni jezik
Engleski
Znanstvena područja
Ekonomija
