Vadlamani Ravi

Web-services classification using intelligent techniques

The web services, a novel paradigm in software technology, have innovative mechanism for rendering services over diversified environment. They promise to allow businesses to adapt rapidly to changes in the business environment and the needs of different customers. The rapid introduction of new web services into a dynamic business environment can adversely affect the service quality and user satisfaction. Consequently, assessment of the quality of web services is of paramount importance in selecting a web service for an application. In this paper, we employed well-known classification models viz., back propagation neural network (BPNN), probabilistic neural network (PNN), group method of data handling (GMDH), classification and regression trees (CART), TreeNet, support vector machine (SVM) and ID3 decision tree (J48) to predict the quality of a web service based on a set of quality attributes. The experiments are carried out on the QWS dataset. We applied 10-fold cross-validation to test the efficacy of the models. The J48 and TreeNet techniques outperformed all other techniques by yielding an average accuracy of 99.72%. We also performed feature selection and found that web-services relevance function (WSRF) is the most significant attribute in determining the quality of a web service. Later, we performed feature selection without WSRF and found that Reliability, Throughput, Successability, Documentation and ResponseTime are the most important attributes in that order. Moreover, the set of if-then rules yielded by J48 and CART can be used as an expert system for web-services classification.

Financial distress prediction in banks using Group Method of Data Handling neural network, counter propagation neural network and fuzzy ARTMAP

This paper presents three hitherto unused neural network architectures for bankruptcy prediction in banks. These networks are Group Method of Data Handling (GMDH), Counter Propagation Neural Network (CPNN) and fuzzy Adaptive Resonance Theory Map (fuzzy ARTMAP). Efficacy of each of these techniques is tested by using four different datasets pertaining to Spanish banks, Turkish banks, UK banks and US banks. Further t-statistic, f-statistic and GMDH are used for feature selection purpose and the features so selected are fed as input to GMDH, CPNN and fuzzy ARTMAP for classification purpose. In each of these cases, top five features are selected in the case of Spanish dataset and top seven features are selected in the case of Turkish and UK datasets. It is observed that the features selected by t-statistic and f-statistic are identical in all datasets. Further, there is a good overlap in the features selected by t-statistic and GMDH. The performance of these hybrids is compared with that of GMDH, CPNN and fuzzy ARTMAP in their stand-alone mode without feature selection. Ten-fold cross validation is performed throughout the study. Results indicate that the GMDH outperformed all the techniques with or without feature selection. Furthermore, the results are much better than those reported in previous studies on the same datasets in terms of average accuracy, average sensitivity and average specificity.

Failure prediction of dotcom companies using neural network-genetic programming hybrids

This paper presents novel neural network-genetic programming hybrids to predict the failure of dotcom companies. These hybrids comprise multilayer feed forward neural network (MLFF), probabilistic neural network (PNN), rough sets (RS) and genetic programming (GP) in a two-phase architecture. In each hybrid, one technique is used to perform feature selection in the first phase and another one is used as a classifier in the second phase. Further t-statistic and f-statistic are also used separately for feature selection in the first phase. In each of these cases, top 10 features are selected and fed to the classifier. Also, the NN-GP hybrids are compared with MLFF, PNN and GP in their stand-alone mode without feature selection. The dataset analyzed here is collected from Wharton Research Data Services (WRDS). It consists of 240 dotcom companies of which 120 are failed and 120 are healthy. Ten-fold cross-validation is performed throughout the study. Results in terms of average accuracy, average sensitivity, average specificity and area under the receiver operating characteristic curve (AUC) indicate that the GP outperformed all the techniques with or without feature selection. The superiority of GP-GP is demonstrated by t-test at 10% level of significance. Furthermore, the results are much better than those reported in previous studies on the same dataset.

Support vector regression based hybrid rule extraction methods for forecasting

Support Vector Regression (SVR) solves regression problems based on the concept of Support Vector Machine (SVM) introduced by Vapnik (1995). The main drawback of these newer techniques is their lack of interpretability. In other words, it is difficult for the human analyst to understand the knowledge learnt by these models during training. The most popular way to overcome this difficulty is to extract if-then rules from SVM and SVR. Rules provide explanation capability to these models and improve the comprehensibility of the system. Over the last decade, different algorithms for extracting rules from SVM have been developed. However rule extraction from SVR is not widely available yet. In this paper a novel hybrid approach for extracting rules from SVR is presented. The proposed hybrid rule extraction procedure has two phases: (1) Obtain the reduced training set in the form of support vectors using SVR (2) Train the machine leaning techniques (with explanation capability) using the reduced training set. Machine learning techniques viz., Classification And Regression Tree (CART), Adaptive Network based Fuzzy Inference System (ANFIS) and Dynamic Evolving Fuzzy Inference System (DENFIS) are used in the phase 2. The proposed hybrid rule extraction procedure is compared to stand-alone CART, ANFIS and DENFIS. Extensive experiments are conducted on five benchmark data sets viz. Auto MPG, Body Fat, Boston Housing, Forest Fires and Pollution, to demonstrate the effectiveness of the proposed approach in generating accurate regression rules. The efficiency of these techniques is measured using Root Mean Squared Error (RMSE). From the results obtained, it is concluded that when the support vectors with the corresponding predicted target values are used, the SVR based hybrids outperform the stand-alone intelligent techniques and also the case when the support vectors with the corresponding actual target values are used.

Software cost estimation using computational intelligence techniques

This paper presents computational intelligence techniques for software cost estimation. We proposed a new recurrent architecture for Genetic Programming (GP) in the process. Three linear ensembles based on (i) arithmetic mean (ii) geometric mean and (iii) harmonic mean are implemented. We also performed GP based feature selection. The efficacy of these techniques viz Multiple Linear Regression, Polynomial Regression, Support Vector Regression, Classification and Regression Tree, Multivariate Adaptive Regression Splines, Multilayer FeedForward Neural Network, Radial Basis Function Neural Network, Counter Propagation Neural Network, Dynamic Evolving Neuro-Fuzzy Inference System, Tree Net, Group Method of Data Handling and Genetic Programming has been tested on the International Software Benchmarking Standards Group (ISBSG) release 10 dataset. Ten-fold cross validation is performed throughout the study. The results obtained from our experiments indicate that new recurrent architecture for Genetic Programming outperformed all the other techniques.

Support vector machine and wavelet neural network hybrid: application to bankruptcy prediction in banks

This paper presents a novel soft computing system, SVWNN, to predict failure of banks. First, support vectors that are critical in classification are extracted from support vector machine (SVM). Then, these support vectors along with their corresponding actual output labels are used to train the wavelet neural network (WNN). Further, Garsons algorithm for feature selection is adapted using WNN. Thus, the new hybrid, WNN-SVWNN, accomplishes horizontal and vertical reduction in the dataset as support vectors reduce the pattern space dimension and the WNN-based feature selection reduces the feature space dimension. The effectiveness of these hybrids is demonstrated on the datasets of US, Turkish, UK and Spanish banks. SVWNN outperformed SVM and WNN on all datasets except Spanish banks. However, when feature selection is considered, WNN-SVM outperformed WNN-WNN and WNN-SVWNN on Spanish and Turkish banks, while WNN-SVWNN outscored others on UK banks. Ten-fold cross-validation was performed throughout the study.

A hybrid under-sampling approach for mining unbalanced datasets: applications to banking and insurance

In solving unbalanced classification problems, machine learning algorithms are overwhelmed by the majority class and consequently misclassify the minority class observations. Here, we propose a hybrid under-sampling approach to improve the performance of classifiers. The proposed approach first employs k-reverse nearest neighbour (kRNN) method to detect the outliers from majority class. After removing the outliers, using K-means clustering, K-clusters are selected to further reduce the influence of the majority class. Then, we employed support vector machine (SVM), logistic regression (LR), multi layer perceptron (MLP), radial basis function network (RBF), group method of data handling (GMDH), genetic programming (GP) and decision tree (J48) for classification purpose. The effectiveness of the proposed approach was demonstrated on datasets taken from insurance fraud detection and credit card churn in banking domain. Ten-fold cross validation method was used in the study. It is observed that the proposed approach improved the performance of the classifiers.

Data Mining Using Rules Extracted from SVM: An Application to Churn Prediction in Bank Credit Cards

In this work, an eclectic procedure for rule extraction from Support Vector Machine is proposed, where Tree is generated using Naive Bayes Tree (NBTree) resulting in the SVM+NBTree hybrid. The data set analyzed in this paper is about churn prediction in bank credit cards and is obtained from Business Intelligence Cup 2004. The data set under consideration is highly unbalanced with 93.11% loyal and 6.89% churned customers. Since identifying churner is of paramount importance from business perspective, sensitivity of classification model is more critical. Using the available, original unbalanced data only, we observed that the proposed hybrid SVM+NBTree yielded the best sensitivity compared to other classifiers.

Software Reliability Prediction Using Group Method of Data Handling

The main purpose of this paper is to propose the use of Group Method of Data Handling (GMDH) to predict software reliability. The GMDH algorithm presented in this paper is a heuristic self-organization method. It establishes the input-output relationship of a complex system using multilayered perception type structure that is similar to a feed forward multilayer neural network. The effectiveness of GMDH is demonstrated on a dataset taken from literature. Its performance is compared with that of multiple linear regression (MLR), back propagation trained neural networks (BPNN), threshold accepting trained neural network (TANN), general regression neural network (GRNN), pi-sigma network (PSN), dynamic evolving neuro-fuzzy inference system (DENFIS), TreeNet, multivariate adaptive regression splines (MARS) and wavelet neural network (WNN) in terms of normalized root mean square error (NRMSE). Based on experiments conducted, it is found that GMDH predicted reliability with least error compared to other techniques. Hence, GMDH can be used a sound alternative to the existing techniques for software reliability prediction.

Failure prediction of banks using threshold accepting trained kernel principal component neural network

This paper presents a new neural network architecture kernel principal component neural network (KPCNN) trained by threshold accepting based training algorithm with different kernels like polynomial, sigmoid and Gaussian and its application to bankruptcy prediction in banks. KPCNN is a non linear version of the PCNN proposed elsewhere. In this architecture, dimensionality reduction is taken care of kernel principal component analysis. First the kernel matrices are computed and then PCNN is applied to those kernel matrices. The nonlinearity is introduced into the architecture by applying different kernels like polynomial, sigmoid and Gaussian etc. The efficiency of KPCNN is tested on different datasets including, Spanish banks, Turkish banks and UK banks datasets. Further t-statistic and f-statistic are used for feature selection purpose and the features so selected are fed as input to KPCNN for classification purpose It is observed that the features selected by t-statistic and f-statistic are identical in all datasets. Ten-fold cross validation is performed throughout the study. The performance of KPCNN on above datasets is compared with that of earlier results both with and without feature selection. From this study we can conclude that the KPCNN yields comparable results with all the techniques both with and without feature selection. Furthermore, we can conclude that this KPCNN best suits for the datasets with high nonlinearity.