Combining spreadsheet smells for improved fault prediction

Abstract

Spreadsheets are commonly used in organizations as a programming tool for business-related calculations and decision making. Since faults in spreadsheets can have severe business impacts, a number of approaches from general software engineering have been applied to spreadsheets in recent years, among them the concept of code smells. Smells can in particular be used for the task of fault prediction. An analysis of existing spreadsheet smells, however, revealed that the predictive power of individual smells can be limited. In this work we therefore propose a machine learning based approach which combines the predictions of individual smells by using an AdaBoost ensemble classi er. Experiments on two public datasets containing real-world spreadsheet faults show signi cant improvements in terms of fault prediction accuracy.

Metric-Based Fault Prediction for Spreadsheets

“…To provide a point of reference for the proposed approach, we compared the evaluation results with our own previous work and with related efforts by other researchers. In comparison to our previous work in which our models were based on a smaller catalog of smell metrics [1], using the extensive metric catalog proposed in our current work led to large improvements in terms of prediction performance. In particular, using the new catalog in combination with an AB model led to an improvement of the F1 value from 43 % to 95 % on the Enron Errors dataset.…”

“…Like any other type of software, spreadsheets can contain faults leading to substantial economic losses. 1 Numerous academic approaches for spreadsheet fault prevention, detection, localization, and repair have been proposed [3]. The effectiveness of such approaches often depends on their ability to predict the likelihood that a certain formula of a spreadsheet is faulty.…”

“…One of the first successful applications of smell detection techniques for fault prediction in spreadsheets is a debugging algorithm that uses detected smells [15]. In our own recent work [1], we investigated the use of spreadsheet smells to train an ML-based fault predictor for spreadsheets. However, the obtained results were preliminary, as we focused on a limited set of spreadsheet metrics and we did not conclusively investigate the choice of an appropriate learning technique.…”