Probabilistic Modeling and Visualization for Bankruptcy Prediction

Francisco Antunes, Bernardete Ribeiro, Francisco Camara Pereira

    Research output: Contribution to journalJournal articleResearchpeer-review

    46 Downloads (Pure)

    Abstract

    In accounting and finance domains, bankruptcy prediction is of great utility for all of the economic stakeholders. The challenge of accurate assessment of business failure prediction, specially under scenarios of financial crisis, is known to be complicated. Although there have been many successful studies on bankruptcy detection, seldom probabilistic approaches were carried out. In this paper we assume a probabilistic point-of-view by applying Gaussian Processes (GP) in the context of bankruptcy prediction, comparing it against the Support Vector Machines (SVM) and the Logistic Regression (LR). Using real-world bankruptcy data, an in-depth analysis is conducted showing that, in addition to a probabilistic interpretation, the GP can effectively improve the bankruptcy prediction performance with high accuracy when compared to the other approaches. We additionally generate a complete graphical visualization to improve our understanding of the different attained performances, effectively compiling all the conducted experiments in a meaningful way. We complete our study with an entropy-based analysis that highlights the uncertainty handling properties provided by the GP, crucial for prediction tasks under extremely competitive and volatile business environments.
    Original languageEnglish
    JournalApplied Soft Computing
    Volume60
    Pages (from-to)831-843
    ISSN1568-4946
    DOIs
    Publication statusPublished - 2017

    Cite this

    Antunes, Francisco ; Ribeiro, Bernardete ; Pereira, Francisco Camara. / Probabilistic Modeling and Visualization for Bankruptcy Prediction. In: Applied Soft Computing. 2017 ; Vol. 60. pp. 831-843.
    @article{55670fb9c9e541bc92ae3af283e7d86e,
    title = "Probabilistic Modeling and Visualization for Bankruptcy Prediction",
    abstract = "In accounting and finance domains, bankruptcy prediction is of great utility for all of the economic stakeholders. The challenge of accurate assessment of business failure prediction, specially under scenarios of financial crisis, is known to be complicated. Although there have been many successful studies on bankruptcy detection, seldom probabilistic approaches were carried out. In this paper we assume a probabilistic point-of-view by applying Gaussian Processes (GP) in the context of bankruptcy prediction, comparing it against the Support Vector Machines (SVM) and the Logistic Regression (LR). Using real-world bankruptcy data, an in-depth analysis is conducted showing that, in addition to a probabilistic interpretation, the GP can effectively improve the bankruptcy prediction performance with high accuracy when compared to the other approaches. We additionally generate a complete graphical visualization to improve our understanding of the different attained performances, effectively compiling all the conducted experiments in a meaningful way. We complete our study with an entropy-based analysis that highlights the uncertainty handling properties provided by the GP, crucial for prediction tasks under extremely competitive and volatile business environments.",
    author = "Francisco Antunes and Bernardete Ribeiro and Pereira, {Francisco Camara}",
    year = "2017",
    doi = "10.1016/j.asoc.2017.06.043",
    language = "English",
    volume = "60",
    pages = "831--843",
    journal = "Applied Soft Computing",
    issn = "1568-4946",
    publisher = "Elsevier",

    }

    Probabilistic Modeling and Visualization for Bankruptcy Prediction. / Antunes, Francisco; Ribeiro, Bernardete; Pereira, Francisco Camara.

    In: Applied Soft Computing, Vol. 60, 2017, p. 831-843.

    Research output: Contribution to journalJournal articleResearchpeer-review

    TY - JOUR

    T1 - Probabilistic Modeling and Visualization for Bankruptcy Prediction

    AU - Antunes, Francisco

    AU - Ribeiro, Bernardete

    AU - Pereira, Francisco Camara

    PY - 2017

    Y1 - 2017

    N2 - In accounting and finance domains, bankruptcy prediction is of great utility for all of the economic stakeholders. The challenge of accurate assessment of business failure prediction, specially under scenarios of financial crisis, is known to be complicated. Although there have been many successful studies on bankruptcy detection, seldom probabilistic approaches were carried out. In this paper we assume a probabilistic point-of-view by applying Gaussian Processes (GP) in the context of bankruptcy prediction, comparing it against the Support Vector Machines (SVM) and the Logistic Regression (LR). Using real-world bankruptcy data, an in-depth analysis is conducted showing that, in addition to a probabilistic interpretation, the GP can effectively improve the bankruptcy prediction performance with high accuracy when compared to the other approaches. We additionally generate a complete graphical visualization to improve our understanding of the different attained performances, effectively compiling all the conducted experiments in a meaningful way. We complete our study with an entropy-based analysis that highlights the uncertainty handling properties provided by the GP, crucial for prediction tasks under extremely competitive and volatile business environments.

    AB - In accounting and finance domains, bankruptcy prediction is of great utility for all of the economic stakeholders. The challenge of accurate assessment of business failure prediction, specially under scenarios of financial crisis, is known to be complicated. Although there have been many successful studies on bankruptcy detection, seldom probabilistic approaches were carried out. In this paper we assume a probabilistic point-of-view by applying Gaussian Processes (GP) in the context of bankruptcy prediction, comparing it against the Support Vector Machines (SVM) and the Logistic Regression (LR). Using real-world bankruptcy data, an in-depth analysis is conducted showing that, in addition to a probabilistic interpretation, the GP can effectively improve the bankruptcy prediction performance with high accuracy when compared to the other approaches. We additionally generate a complete graphical visualization to improve our understanding of the different attained performances, effectively compiling all the conducted experiments in a meaningful way. We complete our study with an entropy-based analysis that highlights the uncertainty handling properties provided by the GP, crucial for prediction tasks under extremely competitive and volatile business environments.

    U2 - 10.1016/j.asoc.2017.06.043

    DO - 10.1016/j.asoc.2017.06.043

    M3 - Journal article

    VL - 60

    SP - 831

    EP - 843

    JO - Applied Soft Computing

    JF - Applied Soft Computing

    SN - 1568-4946

    ER -