Predictive accuracy comparison between neural networks and statistical regression for development effort of software projects

To get a better prediction of costs, schedule, and the risks of a software project, it is necessary to have a more accurate prediction of its development effort. Among the main prediction techniques are those based on mathematical models, such as statistical regressions or machine learning (ML). The ML models applied to predicting the development effort have mainly based their conclusions on the following weaknesses: (1) using an accuracy criterion which leads to asymmetry, (2) applying a validation method that causes a conclusion instability by randomly selecting the samples for training and testing the models, (3) omitting the explanation of how the parameters for the neural networks were determined, (4) generating conclusions from models that were not trained and tested from mutually exclusive data sets, (5) omitting an analysis of the dependence, variance and normality of data for selecting the suitable statistical test for comparing the accuracies among models, and (6) reporting results without showing a statistically significant difference. In this study, these six issues are addressed when comparing the prediction accuracy of a radial Basis Function Neural Network (RBFNN) with that of a regression statistical (the model most frequently compared with ML models), to feedforward multilayer perceptron (MLP, the most commonly used in the effort prediction of software projects), and to general regression neural network (GRNN, a RBFNN variant). The hypothesis tested is the following: the accuracy of effort prediction for RBFNN is statistically better than the accuracy obtained from a simple linear regression (SLR), MLP and GRNN when adjusted function points data, obtained from software projects, is used as the independent variable. Samples obtained from the International Software Benchmarking Standards Group (ISBSG) Release 11 related to new and enhanced projects were used. The models were trained and tested from a leave-one-out cross-validation method. The criteria for evaluating the models were based on Absolute Residuals and by a Friedman statistical test. The results showed that there was a statistically significant difference in the accuracy among the four models for new projects, but not for enhanced projects. Regarding new projects, the accuracy for RBFNN was better than for a SLR at the 99% confidence level, whereas the MLP and GRNN were better than for a SLR at the 90% confidence level.     

Predicting the type of pregnancy using artificial neural networks and multinomial logistic regression: a comparison study

Some medical and epidemiological surveys have been designed to predict a nominal response variable with several levels. With regard to the type of pregnancy there are four possible states: wanted, unwanted by wife, unwanted by husband and unwanted by couple. In this paper, we have predicted the type of pregnancy, as well as the factors influencing it using two different models and comparing them. Regarding the type of pregnancy with several levels, we developed a multinomial logistic regression and a neural network based on the data and compared their results using three statistical indices: sensitivity, specificity and kappa coefficient. Based on these three indices, neural network proved to be a better fit for prediction on data in comparison to multinomial logistic regression. When the relations among variables are complex, one can use neural networks instead of multinomial logistic regression to predict the nominal response variables with several levels in order to gain more accurate predictions.     

A comparison of neural network and multiple regression analysis in modeling capital structure

Empirical studies of the variation in debt ratios across firms have used statistical models singularly to analyze the important determinants of capital structure. Researchers, however, rarely employ non-linear models to examine the determinants and make little effort to identify a superior prediction model. This study adopts multiple linear regressions and artificial neural networks (ANN) models with seven explanatory variables of corporation’s feature and three external macro-economic control variables to analyze the important determinants of capital structures of the high-tech and traditional industries in Taiwan, respectively. Results of this study show that the determinants of capital structure are different in both industries. The major different determinants are business-risk and growth opportunities. Based on the values of RMSE, the ANN models achieve a better fit and forecast than the regression models for debt ratio, and ANNs are cable of catching sophisticated non-linear integrating effects in both industries. It seems that the relationships between debt ratio and independent variables are not linear. Managers can apply these results for their dynamic adjustment of capital structure in achieving optimality and maximizing firm’s value.     

Electrochemical performances of B doped and undoped diamond-like carbon (DLC) films deposited by femtosecond pulsed laser ablation for heavy metal detection using square wave anodic stripping voltammetric (SWASV) technique

Pure diamond-like carbon (DLC) thin films and boron-doped DLC thin films have been deposited on silicon substrates using femtosecond pulsed laser. The amorphous carbon materials (DLC), have been deposited at room temperature by ablating graphite targets with an amplified Ti:sapphire laser of 800 nm wavelength and a pulse duration of 150 fs in high vacuum conditions. Doping with boron has been performed by ablating alternatively graphite and boron targets.In this study, the DLC films were used as working electrodes for the electrochemical detection of trace heavy metals namely, Cd²⁺, Pb²⁺, Ni²⁺ and Hg²⁺, by using square wave anodic stripping voltammetry (SWASV) technique. Four metals were detected at −1.3 V deposition potential, and 90 s deposition time. The DLC films have been characterized by multiwavelength Raman spectrometry and high resolution scanning electron microscopy. The effect of the boron doping on the electrochemical behavior has been shown. The a-C:B 8%/Si₃N₄ electrode gives the more sensitive detection. The four metals are detected simultaneously with a detection limit of 1 μg/L or 2 μg/L and a dynamic range from 1 or 2 to 25 μg/L for every metal, as presented in third table of this article. The different sensitivities obtained are 6.2, 20.0, 1.2 and 6.6 μA/ppb or μA μg⁻¹ L for Cd²⁺, Pb²⁺, Ni²⁺ and Hg²⁺, respectively.     

A study of academic performance of business school graduates using neural network and statistical techniques

Over the past several years, there is tremendous increase in the number of applicants to business schools and hence adequately measuring the potential of these students with regard to their academic performance is an important process of admission decision for any business school. In the present study, an analysis is carried out to predict the academic performance of business school graduates using neural networks and traditional statistical techniques and results are compared to evaluate the performance of these techniques. The underlying constructs in a traditional business school curriculum are also identified and its relevance with the various elements of admission process is presented.     

A comparison of genetic programming and artificial neural networks in metamodeling of discrete-event simulation models

Genetic programming (GP) and artificial neural networks (ANNs) can be used in the development of surrogate models of complex systems. The purpose of this paper is to provide a comparative analysis of GP and ANNs for metamodeling of discrete-event simulation (DES) models. Three stochastic industrial systems are empirically studied: an automated material handling system (AMHS) in semiconductor manufacturing, an (s,S) inventory model and a serial production line. The results of the study show that GP provides greater accuracy in validation tests, demonstrating a better generalization capability than ANN. However, GP when compared to ANN requires more computation in metamodel development. Even given this increased computational requirement, the results presented indicate that GP is very competitive in metamodeling of DES models.     

Segmentation of malignant tumours in mammogram images: A hybrid approach using convolutional neural networks and connected component analysis

The segmentation of breast lesions is an important step in the computer-aided analysis of the mammogram. The presence of noise in mammograms makes lesion detection challenging particularly for complex malignant lesions. Pre-processing techniques can deal with the noise issue but distorts the important shape features. This motivates us to propose a novel hybrid approach by combining a convolution neural network (CNN) with connected component analysis (CCA) to segment malignant breast lesions without any pre-processing to avoid any distortion in image sharpness at the initial stages. Two well-known segmentation techniques namely, K-means (KM) and Fuzzy c-means (FCM) are also used to compare the results. From a pool of 1045 mammographic cancer images acquired from the Digital Database for Screening Mammography (DDSM), 1016 are used for training and validation, and 29 are used for testing. All three results (Hybrid, KM and FCM) are compared against the results by the expert Radiologist. The results indicate that, among various segmentation techniques, the proposed hybrid approach achieves the highest accuracy (90%), Matthew's correlation coefficient (0.79), Jaccard index (0.73) and the Dice similarity coefficient (0.84). Other performance evaluation techniques such as; precision, sensitivity, specificity, false-positive rate, false discovery rate, negative predictive value and false-negative rate also show the superior performance of the proposed hybrid approach. Statistical analysis (Mann–Whitney U test, T-test, Chi-square test, Kolmogorov–Smirnov test and Wilcoxon test), graphical analysis (Regression and Bland–Altman plots) and receiver operating characteristic curve further demonstrate the stability and consistency of the results.     

Sensitivity analysis of artificial neural networks for just-suspension speed prediction in solid-liquid mixing systems: Performance comparison of MLPNN and RBFNN

Just-suspension speed (Njs) is an important parameter for stirred tank design using a solid-liquid mixing system in the chemical process industry. However, current correlations for Njs suffer from uncertainty from limited experimental databases and limitations due to many parameters that play an important role in Njs determination. A comprehensive computation of the radial basis function neural network (RBFNN) was developed based on solid-liquid mixing experiments, which contain 935 datasets for the prediction of Njs. The Njs values were obtained experimentally using Zwietering correlation with different solid loading percentages, solid particle density, solid particle diameter, mixing solvent density, number of impeller blades, impeller diameter, impeller blade hub angle, impeller blade tip angle, the width of the impeller blade and the ratio of the clearance between the impeller and the bottom of the tank with the tank diameter. The RBFNN proved to have a much better ability to accurately predict the desired Njs compared to MLPNN even after decreasing the number of input variables from 11 to 8. Thus, the computational RBFNN model results will be useful for extending the application of a solid-liquid mixing system for estimating the just-suspension speed for stirred tank design.