Software Development Metrics Prediction Using Time Series Methods

Abstract

The software development process is an intricate task, with the growing complexity of software solutions and inflating code-line count being part of the reason for the fall of software code coherence and readability thus being one of the causes for software faults and it's declining quality. Debugging software during development is significantly less expensive than attempting damage control after the software's release. An automated quality-related analysis of developed code, which includes code analysis and correlation of development data like an ideal solution. In this paper the ability to predict software faults and software quality is scrutinized. Hereby we investigate four models that can be used to analyze time-based data series for prediction of trends observed in the software development process are investigated. Those models are Exponential Smoothing, the Holt-Winters Model, Autoregressive Integrated Moving Average (ARIMA) and Recurrent Neural Networks (RNN). Time-series analysis methods prove a good fit for software related data prediction. Such methods and tools can lend a helping hand for Product Owners in their daily decision-making process as related to e.g. assignment of tasks, time predictions, bugs predictions, time to release etc. Results of the research are presented.

Data-Driven Requirements Engineering: A Guided Tour

“… Analysis of trends and summarization of results, to understand the direction of a software project.  Prediction of the evolution of a particular indicator or some of the quality factors used to compute it [64].  Definition and triggering of alerts, to report underperformance or (in combined use with prediction) to anticipate future threshold violations.…”

Predicting the Number of Reported Bugs in a Software Repository

“…4. ARIMA: It is one of the most popular models used for time series prediction [2,14,15,24]. The general ARIMA model (p, q, d) is formulated as…”

“…Time series prediction models have been used in the software engineering domain for the past 20 years. Choraś et al [2] exploit time series methods such as ARIMA, Holt-Winters, and random walk to forecast various project-related characteristics. Pati and Shukla [15] compared ARIMA and Artificial Neural Network models for Debian Bug Number Prediction.…”