Model interpretation is one of the key aspects of the model evaluation process. The explanation of the relationship between model variables and outputs is relatively easy for statistical models, such as linear regressions, thanks to the availability of model parameters and their statistical significance. For "black box" models, such as random forest, this information is hidden inside the model structure. This work presents an approach for computing feature contributions for random forest classification models. It allows for the determination of the influence of each variable on the model prediction for an individual instance. By analysing feature contributions for a training dataset, the most significant variables can be determined and their typical contribution towards predictions made for individual classes, i.e., class-specific feature contribution "patterns", are discovered. These patterns represent a standard behaviour of the model and allow for an additional assessment of the model reliability for a new data. Interpretation of feature contributions for two UCI benchmark datasets shows the potential of the proposed methodology. The robustness of results is demonstrated through an extensive analysis of feature contributions calculated for a large number of generated random forest models. * a.m.wojak@bradford.ac.uk

Article:Azzeh, Mohammad, Neagu, Daniel and Cowling, Peter I. orcid.org/0000-0003-1310-6683 (2010) Fuzzy grey relational analysis for software effort estimation. Empirical Software Engineering. pp. 60-90. ISSN 1382-3256 https://doi.org/10.1007/s10664-009-9113-0 eprints@whiterose.ac.uk https://eprints.whiterose.ac.uk/ Reuse Items deposited in White Rose Research Online are protected by copyright, with all rights reserved unless indicated otherwise. They may be downloaded and/or printed for private study, or other acts as permitted by national copyright laws. The publisher or other rights holders may allow further reproduction and re-use of the full text version. This is indicated by the licence information on the White Rose Research Online record for the item. TakedownIf you consider content in White Rose Research Online to be in breach of UK law, please notify us by emailing eprints@whiterose.ac.uk including the URL of the record and the reason for the withdrawal request. Fuzzy Grey Relational Analysis for Software Effort Estimation AbstractAccurate and credible software effort estimation is a challenge for academic research and software industry. From many software effort estimation models in existence, Estimation by Analogy (EA) is still one of the preferred techniques by software engineering practitioners because it mimics the human problem solving approach. Accuracy of such a model depends on the characteristics of the dataset, which is subject to considerable uncertainty. The inherent uncertainty in software attribute measurement has significant impact on estimation accuracy because these attributes are measured based on human judgment and are often vague and imprecise. To overcome this challenge we propose a new formal EA model based on the integration of Fuzzy set theory with Grey Relational Analysis (GRA). Fuzzy set theory is employed to reduce uncertainty in distance measure between two tuples at the k).GRA is a problem solving method that is used to assess the similarity between two tuples with M features. Since some of these features are not necessary to be continuous and may have nominal and ordinal scale type, aggregating different forms of similarity measures will increase uncertainty in the similarity degree. Thus the GRA is mainly used to reduce uncertainty in the distance measure between two software projects for both continuous and categorical features. Both techniques are suitable when relationship between effort and other effort drivers is complex. Experimental results showed that using integration of GRA with FL produced credible estimates when compared with the results obtained using Case-Based Reasoning, Multiple LinearRegression and Artificial Neural Networks methods.

Model interpretation is one of the key aspects of the model evaluation process. The explanation of the relationship between model variables and outputs is easy for statistical models, such as linear regressions, thanks to the availability of model parameters and their statistical significance. For "black box" models, such as random forest, this information is hidden inside the model structure. This work presents an approach for computing feature contributions for random forest classification models. It allows for the determination of the influence of each variable on the model prediction for an individual instance. Interpretation of feature contributions for two UCI benchmark datasets shows the potential of the proposed methodology. The robustness of results is demonstrated through an extensive analysis of feature contributions calculated for a large number of generated random forest models.

No abstract

BackgroundDue to recent advances in data storage and sharing for further data processing in predictive toxicology, there is an increasing need for flexible data representations, secure and consistent data curation and automated data quality checking. Toxicity prediction involves multidisciplinary data. There are hundreds of collections of chemical, biological and toxicological data that are widely dispersed, mostly in the open literature, professional research bodies and commercial companies. In order to better manage and make full use of such large amount of toxicity data, there is a trend to develop functionalities aiming towards data governance in predictive toxicology to formalise a set of processes to guarantee high data quality and better data management. In this paper, data quality mainly refers in a data storage sense (e.g. accuracy, completeness and integrity) and not in a toxicological sense (e.g. the quality of experimental results).ResultsThis paper reviews seven widely used predictive toxicology data sources and applications, with a particular focus on their data governance aspects, including: data accuracy, data completeness, data integrity, metadata and its management, data availability and data authorisation. This review reveals the current problems (e.g. lack of systematic and standard measures of data quality) and desirable needs (e.g. better management and further use of captured metadata and the development of flexible multi-level user access authorisation schemas) of predictive toxicology data sources development. The analytical results will help to address a significant gap in toxicology data quality assessment and lead to the development of novel frameworks for predictive toxicology data and model governance.ConclusionsWhile the discussed public data sources are well developed, there nevertheless remain some gaps in the development of a data governance framework to support predictive toxicology. In this paper, data governance is identified as the new challenge in predictive toxicology, and a good use of it may provide a promising framework for developing high quality and easy accessible toxicity data repositories. This paper also identifies important research directions that require further investigation in this area.