Exploring Preference of Chronological and Relevancy Filtering in Effort Estimation

摘要

BACKGROUND: Effort estimation models are often built based on history data from past projects in an organization. Filtering techniques have been proposed for improving the estimation accuracy. Chronological filtering relies on the time proximity among project data and ignores much old data. Relevancy filtering utilizes the proximity of characteristics among project data and ignores dissimilar data. Their interaction is interesting because one would be able to make more accurate estimates if a positive synergistic effect exists. AIMS: To examine whether the chronological filtering and the relevancy filtering can contribute to improving the estimation accuracy together. METHOD: moving windows approaches as chronological filtering and a nearest neighbor approach as relevancy filtering are applied to a single-company ISBSG data. RESULTS: we observed a negative synergistic effect. Each of the filtering approaches brought better effort estimates than using the whole history data. However, their combination may cause worse effort estimates than using the whole history data. CONCLUSIONS: Practitioners should care about a negative synergistic effect when combining the chronological filtering and the relevancy filtering.