Anatoli Nachev

Fuzzy, Distributed, Instance Counting, And Default Artmap Neural Networks For Financial Diagnosis

This paper shows the potential of neural networks based on the Adaptive Resonance Theory as tools that generate warning signals when bankruptcy of a company is expected (bankruptcy prediction problem). Using that class of neural networks is still unexplored to date. We examined four of the most popular networks of the class — fuzzy, distributed, instance counting, and default ARTMAP. In order to illustrate their performance and to compare with other techniques, we used data, financial ratios, and experimental conditions identical to those published in previous studies. Our experiments show that two financial ratios provide highest discriminatory power of the model and ensure best prediction accuracy. We examined performance and validated results by exhaustive search of input variables, cross-validation, receiver operating characteristic analysis, and area under curve metric. We also did application-specific cost analysis. Our results show that distributed ARTMAP outperforms the other three models in general, but the fuzzy model is best performer for certain vigilance values and in the application-specific context. We also found that ARTMAP outperforms the most popular neural networks — multi-layer perceptrons and other statistical techniques applied to the same data.

Fuzzy ARTMAP Neural Network for Classifying the Financial Health of a Firm

In this paper, an application, based on data from a popular dataset, shows in an empirical form the strengths and weaknesses of fuzzy ARTMAP neural networks as predictor of corporate bankruptcy. This is an advantageous approach enabling fast learning, self-determination of the network structure and high prediction accuracy. Experiments showed that the fuzzy ARTMAP outperforms statistical techniques and the most popular backpropagation MLP neural networks, all applied to the same dataset. An exhaustive search procedure over the Altmans financial ratios leads to the conclusion that two of them are enough to obtain the highest prediction accuracy. The experiments also showed that the model is not sensitive to outliers of the dataset. Our research is the first to use fuzzy ARTMAP neural networks for bankruptcy prediction.

Insolvency Prediction of Irish Companies Using Backpropagation and Fuzzy ARTMAP Neural Networks

This study explores experimentally the potential of BPNNs and Fuzzy ARTMAP neural networks to predict insolvency of Irish firms. We used financial information for Irish companies for a period of six years, preprocessed properly in order to be used with neural networks. Prediction results show that with certain network parameters the Fuzzy ARTMAP model outperforms BPNN. It outperforms also self-organising feature maps as reported by other studies that use the same dataset. Accuracy of predictions was validated by ROC analysis, AUC metrics, and leave-one-out cross-validation.

Comparative analysis of distributed, default, IC, and fuzzy ARTMAP neural networks for classification of malignant and benign lesions

Only one third of all breast cancer biopsies made today confirm the disease, which make these procedures inefficient and expensive. We address the problem by exploring and comparing characteristics of four neural networks used as predictors: fuzzy, distributed, default, and ic ARTMAP, all based on the adaptive resonance theory. The networks were trained using a dataset that contains a combination of 39 mammographic, sonographic, and other descriptors, which is novel for the field. We compared the model performances by using ROC analysis and metrics derived from it, such as max accuracy, full and partial area under the convex hull, and specificity at 98% sensitivity. Our findings show that the four models outperform the most popular MLP neural networks given that they are setup properly and used with appropriate selection of data variables. We also find that two of the models, distributed and ic, are too conservative in their predictions and do not provide sufficient sensitivity and specificity, but the default ARTMAP shows very good characteristics. It outperforms not only its counterparts, but also all other models used with the same data, even some radiologist practices. To the best of our knowledge, the ARTMAP neural networks have not been studied for the purpose of the task until now.

Support Vector Machines for Insolvency Prediction of Irish Companies

This study explores experimentally the potential of linear and non-linear support vector machines with three kernels to predict insolvency of Irish firms. The dataset used contains selected financial features based on information collected from 88 companies for a period of six years. Experiments show that non-linear support vector machines (SVM) with polynomial kernel gives highest prediction accuracy and outperforms all other techniques used so far with the same dataset. SVM performance is estimated by various metrics, receiver operating characteristics analysis, and results are validated by the leave-one-out cross-validation technique.