Gavin Finnie

A comparison of software effort estimation techniques: using function points with neural networks, case-based reasoning and regression models

Estimating software development effort remains a complex problem attracting considerable research attention. Improving the estimation techniques available to project managers would facilitate more effective control of time and budgets in software development. This paper reviews a research study comparing three estimation techniques using function points as an estimate of system size. The models considered are based on regression analysis, artificial neural networks and case-based reasoning. Although regression models performed poorly on the data set of 299 projects, both artificial neural networks and case-based reasoning appeared to have value for software development effort estimation models. Case-based reasoning in particular is appealing because of its similarity to expert judgment approaches and for its potential as an expert assistant in support of human judgment.

Estimating software development effort with connectionist models

Abstract Accurate software development effort estimation is important for effective project management. Research studies indicate that effort estimation is a complex issue and results have in general not been encouraging. Artificial Neural Networks are recognised for their ability to provide good results when dealing with problems where there are complex relationships between inputs and outputs, and where the input data is distorted by high noise levels. This paper reports on the assessment of back-propagation neural network models for effort estimation. The models were tested on simulated data as well as actual data of commercial projects. This project data had large productivity variations, noise and missing data values, which enabled model evaluation under typical software development conditions. The results were encouraging, with the networks showing an ability to estimate development effort within 25% of actual effort more than 75% of the time for one large commercial data set.

R 5 model for case-based reasoning

This paper reviews some existing models of case-based reasoning (CBR) such as the R4 model of CBR and proposes a R5 model, in which repartition, retrieve, reuse, revise and retain are the main tasks for the CBR process. The original idea behind this model is that case base building is an important part of CBR and the case base can be built based on partitioning of the possible world of problems and solutions. It argues that the proposed R5 model is a new approach to using similarity-based reasoning to unify case base building, case retrieval, and case adaptation, and therefore facilitates the development of CBR with applications.

An empirical methodology for developing stockmarket trading systems using artificial neural networks

A great deal of work has been published over the past decade on the application of neural networks to stockmarket trading. Individual researchers have developed their own techniques for designing and testing these neural networks, and this presents a difficulty when trying to learn lessons and compare results. This paper aims to present a methodology for designing robust mechanical trading systems using soft computing technologies, such as artificial neural networks. This paper describes the key steps involved in creating a neural network for use in stockmarket trading, and places particular emphasis on designing these steps to suit the real-world constraints the neural network will eventually operate in. Such a common methodology brings with it a transparency and clarity that should ensure that previously published results are both reliable and reusable.

Similarity and metrics in case‐based reasoning

Similarity is a core concept in case‐based reasoning (CBR), because case base building, case retrieval, and even case adaptation all use similarity or similarity‐based reasoning. However, there is some confusion using similarity, similarity measures, and similarity metrics in CBR, in particular in domain‐dependent CBR systems. This article attempts to resolve this confusion by providing a unified framework for similarity, similarity relations, similarity measures, and similarity metrics, and their relationship. This article also extends some of the well‐known results in the theory of relations to similarity metrics. It appears that such extension may be of significance in case base building and case retrieval in CBR, as well as in various applied areas in which similarity plays an important role in system behavior.

Estimating Software Development Effort with Case-Based Reasoning

Software project effort estimation is a difficult problem complicated by a variety of interrelated factors. Current regression-based models have not had much success in accurately estimating system size. This paper describes a case based reasoning approach to software estimation which performs somewhat better than regression models based on the same data and which has some similarity to human expert judgement approaches. An analysis is performed to determine whether different forms of averaging and adaptation improve the overall quality of the estimate.

Evaluating usability, user satisfaction and intention to revisit for successful e-government websites

This paper determines a set of usability factors for evaluating e-government websites and describes causal effects, which determine the extent to which e-government website usability affects user satisfaction and their intention to revisit sites for continued usage. Measurement data was gathered from user testing on the websites of representative administration departments in South Korea. This data was analysed using factor analysis and a structural equation model was developed. Navigation, utilisation of image and graphics, effective readability, utilisation of multimedia technology, site structure and information search capability were shown to be major factors affecting usability of e-government websites. Findings suggest that the usability strongly affected both user satisfaction and intention to revisit.

Experience management in knowledge management

This paper examines experience and knowledge, experience management and knowledge management, and their interrelationships. It also proposes process perspectives for both experience management and knowledge management, which integrate experience processing and corresponding management, knowledge processing and corresponding management respectively. The proposed approach will facilitate research and development of knowledge management and experience management as well as knowledge-based systems.

Case base building with similarity relations

This paper has two main contributions. Firstly, it shows that similarity relations are an adequate means of formalization not only for case retrieval but also for case base building. Secondly, this paper provides a theoretical formalization for building case bases in case-based reasoning and presents three algorithms for case base building. The proposed approach argues that case base building can be based on both similarity relations and fuzzy similarity relations, which are both defined on the possible world of problems and solutions respectively. Thus case base building is a form of similarity-based reasoning. This approach is an extension for the logical and fuzzy approach to case based reasoning. The proposed methods and algorithms can be applied to reduction of case base size.

Enhancing stockmarket trading performance with ANNs

Artificial neural networks (ANNs) have been repeatedly and consistently applied to the domain of trading financial time series, with mixed results. Many researchers have developed their own techniques for both building and testing such ANNs, and this presents a difficulty when trying to learn lessons and compare results. In a previous paper, Vanstone and Finnie have outlined an empirical methodology for creating and testing ANNs for use within stockmarket trading systems. This paper demonstrates the use of their methodology, and creates and benchmarks a financially viable ANN-based trading system. Many researchers appear to fail at the final hurdles in their endeavour to create ANN-based trading systems, most likely due to their lack of understanding of the constraints of real-world trading. This paper also attempts to address this issue.