Forecasting short-term productivity in IT projects
Abstract
Forecasting short-term productivity in IT projects
Incoming article date: 16.01.2018At the moment one of the most advanced flexible methodologies IT project management is SCRUM technology. Its feature is the partition of the whole work on short iterations (sprints), each of which represents a cyclic process, repeated all the way through the product development process. This allows you to constantly monitor the emerging risks and minimize them effects. Each sprint starts with planning, which sets the pace of work and stimulates high performance. The key point of planning is predicting the level of focus of the team on the implementation of the tasks, the so-called focus factor. It is the ratio productivity to labor intensity. The paper presents the results of numerical experiments to identify the best methods for predicting this indicator. In their the production data of one of the project groups of the real commercial firm in the field of IT. Extrapolation forecasting methods with a sliding base on the previous actual results. At the same time, the width of the base one to seven previous periods), and the type of extrapolating function. Approximating polynomials of zero, first and second degree are considered. Attention has also been paid to the method of exponential smoothing and Bayesian approach to diagnosis and prediction. For the integral the quality of forecasting used the value of the root-mean-square relative performance error. According to the results of the experiments, three optimal variants are selected. The first is the forecast for the sliding arithmetic average of the actual focus-factors for 3-4 previous periods. In the second variant, again, the sliding the average of actual focus factors for 3-5 sprints, but not arithmetic, and geometric. The third method for predicting is the product theoretical focus factor on the actual. We use the geometric mean it per the 1-2 sprints. In the mean square norm for both methods, the prediction error is the same, about 5.8%. If we estimate it not by the mean square metric, but by maximum deviation, then the latter option is slightly better. Results are intended to clarify the intuitive planning of sprints when performing real projects.
Keywords: management, IT-project, SCRUM technology, sprint planning, forecast, focus factor, moving average, linear and quadratic regression, exponential smoothing, Bayesian approach, root-mean-square error, the best possible option