K. K. Shukla

Neuro-genetic prediction of software development effort

AbstractPrediction of resource requirements of a software project is crucial for the timely delivery of quality-assured software within a reasonabletimeframe. Many conventional (model-based) and AI-oriented (model-free) resource estimators have been proposed in the recent past. Thispaper presents a novel genetically trained neural network (NN) predictor trained on historical data. We demonstrate substantial improvementin prediction accuracy by the neuro-genetic approach as compared to both a regression-tree-based conventional approach, as well asbackpropagation-trained NN approach reported recently. The superiority of this new predictor is established usingn-fold cross validationand Student’s t-test on various partitions of merged Cocomo and Kemerer data sets incorporating data from 78 real-life software projects.q 2000 Elsevier Science B.V. All rights reserved. Keywords: Neuro-genetic prediction; Neural network predictor; Genetically trained neural network 1. IntroductionReasonably accurate prediction of software developmenteffort has a profound effect on all stages of the softwaredevelopment cycle. Underestimates of resource require-ments for a software project lead to: (a) underestimationof the cost; (b) unrealistic time schedule; (c) considerablework pressure on the engineers; and (d) compromises indevelopment methodology, documentation and testing. Onthe other hand, overestimates are likely to cause: (a) a lostcontract due to prohibitive costs; (b) over allocation of engi-neers to the project leading to constraints on other projects;(c) low productivity levels of engineers; and (d) easy-goingwork habit in the organization. Resource requirementprediction for software projects is, therefore, an activeresearch area.Various conventional model-based methods have metwith limited success, whereas, intelligent prediction usingneurocomputing has proven its worth in many diverse appli-cation areas [1]. McCullagh et al. [2] have used neuralnetwork (NN) to estimate rainfall in Australia and havereported results superior to conventional model-basedapproach. NN predictors are playing major roles in diverseapplications and are being successfully applied to load fore-casting, medical diagnosis, communications, robot naviga-tion, software production etc., for example see Ref. [3].Recently, software engineers have started using NNs invarious stages of software production with significantsuccess. Karunanithi [4] has applied NN for software relia-bility prediction in the presence of code churn. This work isa major step forward in software reliability estimation sincethe conventional reliability growth models made the unrea-listic assumption that the complete code for the system isavailable before testing starts and the code remains frozenduring testing. Due to their power of generalization, NNsare able to accurately predict reliability in the presence ofcode churn. In a unique application of NN-based classifier,Khoshgoftaar et al. [5] have developed a system for identi-fying high-risk, error-prone modules early in the develop-ment cycle to allow optimal resource allocation for themodules. Specification-level software size estimates havebeen obtained by Hakkarainen et al. [6] by training an NNwith structured analysis (SA) descriptions as inputs, and sizemetric values as outputs. The authors used training and testdata set consisting of randomly generated SA descriptionsas input data and corresponding algorithm-based size metricvalues as output data. The size metrics used in their experi-ments were—DeMarco’s Function Bang metric, Albrecht’sFunction Points and Symons’ Mark II Function Points.Function Bang is based on the complexity of data flowsand the types of operation on these data flows. It measuresthe number of data-tokens around the boundary of variousfunctional primitives in a data flow diagram; whereas,

Segmentation and classification of medical images using texture-primitive features: Application of BAM-type artificial neural network

The objective of developing this software is to achieve auto-segmentation and tissue characterization. Therefore, the present algorithm has been designed and developed for analysis of medical images based on hybridization of syntactic and statistical approaches, using artificial neural network (ANN). This algorithm performs segmentation and classification as is done in human vision system, which recognizes objects; perceives depth; identifies different textures, curved surfaces, or a surface inclination by texture information and brightness. The analysis of medical image is directly based on four steps: 1) image filtering, 2) segmentation, 3) feature extraction, and 4) analysis of extracted features by pattern recognition system or classifier. In this paper, an attempt has been made to present an approach for soft tissue characterization utilizing texture-primitive features with ANN as segmentation and classifier tool. The present approach directly combines second, third, and fourth steps into one algorithm. This is a semisupervised approach in which supervision is involved only at the level of defining texture-primitive cell; afterwards, algorithm itself scans the whole image and performs the segmentation and classification in unsupervised mode. The algorithm was first tested on Markov textures, and the success rate achieved in classification was 100%; further, the algorithm was able to give results on the test images impregnated with distorted Markov texture cell. In addition to this, the output also indicated the level of distortion in distorted Markov texture cell as compared to standard Markov texture cell. Finally, algorithm was applied to selected medical images for segmentation and classification. Results were in agreement with those with manual segmentation and were clinically correlated.

Real-time task scheduling with fuzzy uncertainty in processing times and deadlines

In real-time systems, scheduling algorithms play the vital role of devising a feasible schedule of the tasks. The scheduling algorithm designer faces uncertainty associated with the timing constrains of the real-time tasks. This paper considers fuzzy timing constraints by modeling the real-time tasks with fuzzy deadlines and fuzzy processing times with different membership functions. Comparative studies and some interesting findings based on simulation experiments are reported.

Real-time scheduling of periodic tasks with processing times and deadlines as parametric fuzzy numbers

Task scheduling is very important in real-time systems as it accomplishes the crucial goal of devising a feasible schedule of the tasks. However, the uncertainty associated with the timing constrains of the real-time tasks makes the scheduling problem difficult to formulate. This motivates the use of fuzzy numbers to model task deadlines and completion times. In this paper a method for intuitively defining smooth membership functions (MFs) for deadlines and execution times has been proposed using mixed cubic-exponential Hermite interpolation parametric curves. The effect of changes in parameterized MFs on the task schedulability and task priorities are also reported. A new technique is proposed based on the concept of dynamic slack calculation to make the existing model more practical and realistic. Examples are given to demonstrate the more satisfactory performance of the new technique.

Exploring neuro-genetic processing of electronic nose data

This paper explores neuro-genetic applications in processing electronic nose data corrupted with additive Gaussian noise. For this study, published sensor data for different polymer-coated surface-acoustic wave (SAW) sensor arrays exposed to fixed concentrations of hazardous vapours like diethyl sulphide (DES) and iso-octane (ISO) have been used. Dimensionality of resulting pattern recognition problem is varied by taking different numbers of sensors. We show that for low dimensionality instances of this problem, back-propagation performs adequately under noisy conditions. For high dimensionality instances, back-propagation has great difficulty in training the neural classifier even with repeated restarts and different weights initializations. To alleviate this problem, we propose use of a genetic algorithm with special MRX operator introduced by us and demonstrate very encouraging results with a genetically trained neural network model.

Efficient Algorithms for Discrete Wavelet Transform

Wavelet transforms (WT) have growing impact on signal processing theory and practice. This is because of two reasons: (a) unifying role of wavelet transform and (b) their successes in wide variety of applications. Digital filter banks, the basis of wavelet-based algorithms, have become standard signal processing operators. Filter banks are the fundamental tools required for processing of real signals using digital signal processors (DSP) [133,139]. Vaidyanathan in his book [134] has discussed connection between theory of filter bank and DSP. The purpose of this book is to look at wavelet-related issues from a signal processing perspective. This book focuses on and around new implementation techniques of discrete wavelet transform (DWT) and their applications in denoising and classification. On this account, it is required to introduce the wavelet theory in brief. The organization of this chapter is as follows: Section 1.1 introduces the subject in brief. Section 1.2 presents historical review of multiresolution analysis and wavelet transform. Various kinds of wavelet transform applied to signal processing applications viz. continuous wavelet transform (CWT) and DWT (one dimension and two dimensions) are discussed in brief. Section 1.3 reviews implementation issues and applications of DWT from signal processing viewpoint. Section 1.4 concludes this chapter by outlining major contribution of the book.

ADAPTIVE RESONANCE NEURAL CLASSIFIER FOR IDENTIFICATION OF GASES/ODOURS USING AN INTEGRATED SENSOR ARRAY

Abstract A new approach to intelligent gas sensor (IGS) design using a classifier based on adaptive resonance theory (ART) artificial neural network (ANN) is presented. Using published data of sensor arrays fabricated and characterised at our laboratory, we demonstrate excellent gas/odour identification performance of our classifier for a 4-gas, 4-sensor system to identify individual gas/odour. Since the ART neural network is a self-organising classifier trained in the unsupervised mode, it avoids the drawbacks associated with static feedforward neural networks trained with a locally optimal backpropagation-type training algorithms applied by researchers in the recent past. The ART neural network offers easy implementability and real time performance in addition to giving excellent classification accuracy as demonstrated by our experiments.

On the design issue of intelligent electronic nose system

Intelligent electronic nose (ENOSE) system technology is gaining importance in several industrial applications. These include process control and quality control in industries such as foodstuffs, beverages, tobacco, perfumery and pharmaceutical. ENOSE is also crucial component in industrial safety (smoke and hazardous gas detection) as well as environmental pollution control. This paper deals with design of an intelligent ENOSE system for the identification of gas/odours using a sensor array and a neural network pattern classifier. Previous researchers have shown that the power of discrimination increases rapidly with the number of sensors in the array whose information potential is very large and the pattern recognition (PARC) method is a clever way to extract this information. The authors show in this paper with the powerful PARC technique, the need of larger array can be compensated. With this view, they design a neural classifier using two different learning approaches and train the network over the responses of surface acoustic wave (SAW) sensors exposed to hazardous vapours like diethyl sulphide (DES) and iso-octane (ISO). Dimensionality of the data set is varied from 1 to 8 by taking different number of sensors. It is found that for a backpropagation trained neural classifier, the optimum number of sensors required for satisfactory classification under noisy conditions is 4 to 5. This is a very limited range beyond which backpropagation has great difficulty in training the neural classifier even with repeated restarts and different weight initializations. To alleviate this problem, hybridization of soft computing tools like neural networks and genetic algorithms promises to provide the design of better intelligent system. The authors propose the use of a genetic algorithm based on a special MRX operator introduced by them and demonstrate very encouraging results with genetically trained neural network model even with larger as well as smaller numbers of sensors.

Segmentation of medical images using Simulated Annealing Based Fuzzy C Means algorithm

Accurate segmentation is desirable for analysis and diagnosis of medical images. This study provides methodology for fully automated simulated annealing based fuzzy c-means algorithm, modelled as graph search method. The approach is unsupervised based on pixel clustering using textural features. The virtually training free algorithm needs initial temperature and cooling rate as input parameters. Experimentation on more than 180 MR and CT images for different parameter values, has suggested the best-suited values for accurate segmentation. An overall 97% correct segmentation has been achieved. The results, evaluated by radiologists, are of clinical importance for segmentation and classification of Region of Interest.

Passive forensics in image and video using noise features

Due to present of enormous free image and video editing software on the Internet, tampering of digital images and videos have become very easy. Validating the integrity of images or videos and detecting any attempt of forgery without use of active forensic technique such as Digital Signature or Digital Watermark is a big challenge to researchers. Passive forensic techniques, unlike active techniques, do not need any preembeded information about the image or video. The proposed paper presents a comprehensive review of the recent developments in the field of digital image and video forensic using noise features. The previously existing methods of image and video forensics proved the importance of noises and encourage us for the study and perform extensive research in this field. Moreover, in this paper, forensic task cover mainly source identification and forgery detection in the image and video using noise features. Thus, various source identification and forgery detection methods using noise features are reviewed and compared in this paper for image and video. The overall objective of this paper is to give researchers a broad perspective on various aspects of image and video forensics using noise features. Conclusion part of this paper discusses about the importance of noise features and the challenges encountered by different image and video forensic method using noise features.