Figure - uploaded by Rogelio A Mancisidor
Content may be subject to copyright.
![Model performance for 1-year bankruptcy predictions. We report average and standard deviation values for 5 different randomly selected training data sets during the period 1994 to 2007. The test period, in this case, is from 2008 to 2014. Results marked with † are taken from [39].](profile/Rogelio-A-Mancisidor-2/publication/365415689/figure/tbl1/AS:11431281097416680@1668569371497/Model-performance-for-1-year-bankruptcy-predictions-We-report-average-and-standard.png)
Model performance for 1-year bankruptcy predictions. We report average and standard deviation values for 5 different randomly selected training data sets during the period 1994 to 2007. The test period, in this case, is from 2008 to 2014. Results marked with † are taken from [39].
Source publication
This research introduces for the first time the concept of multimodal learning in bankruptcy prediction models. We use the Conditional Multimodal Discriminative (CMMD) model to learn multimodal representations that embed information from accounting, market, and textual modalities. The CMMD model needs a sample with all data modalities for model tra...
Contexts in source publication
Context 1
... means losing about 40% of the companies in our initial sample. Table 4 measures classification performance, based on AUC, for 1 year bankruptcy predictions using the data set for the second set of experiments. LR, SVM, RF, MLP, and CMMD achieve higher AUC values than the DL-Embedding model introduced in [39]. ...Context 2
... test period is always the same and it depends on the forecasting horizon. Training and test periods are shown in Table 4 panel (b). representations based on accounting and market modalities. ...Context 3
... highlights one of the main advantages of our proposed methodology, i.e., CMMD only needs a sample with all data modalities for model training 9 . It is noteworthy that the CMMD classification performance for all 5,203 firms is 0.8919 ± 0.0009, which is very similar to the one reported in Table 4. ...Context 4
... superscripts *, **, and *** denote the best architecture for 1, 2, and 3 years bankruptcy predictions, respectively (see Table 3). Likewise, the subscript **** denotes the best architecture used in the experiments shown in Table 4. For the rest of hyperparamters in the benchmark models, we used default values in the sklearn implementation. ...Context 5
... means losing about 40% of the companies in our initial sample. Table 4 measures classification performance, based on AUC, for 1 year bankruptcy predictions using the data set for the second set of experiments. LR, SVM, RF, MLP, and CMMD achieve higher AUC values than the DL-Embedding model introduced in [39]. ...Context 6
... test period is always the same and it depends on the forecasting horizon. Training and test periods are shown in Table 4 panel (b). representations based on accounting and market modalities. ...Context 7
... highlights one of the main advantages of our proposed methodology, i.e., CMMD only needs a sample with all data modalities for model training 9 . It is noteworthy that the CMMD classification performance for all 5,203 firms is 0.8919 ± 0.0009, which is very similar to the one reported in Table 4. ...Context 8
... superscripts *, **, and *** denote the best architecture for 1, 2, and 3 years bankruptcy predictions, respectively (see Table 3). Likewise, the subscript **** denotes the best architecture used in the experiments shown in Table 4. For the rest of hyperparamters in the benchmark models, we used default values in the sklearn implementation. ...Similar publications
Carrigan and Sidman Journal of the Experimental Analysis of Behavior , 58 , 183-204, (1992) proposed that select and reject arbitrary conditional relations are equivalence relations, each resulting in the emergence of alternative stimulus equivalence classes. The standard matching to sample (MTS) procedure can potentially teach both select and reje...
