A parametric effort estimation model for mobile apps

Abstract

According to a recent report [1], in 2015 alone, the revenue generated from mobile applications is expected to reach $100 billion. This figure indicates that one cannot neglect the potential of mobile application development business. Yet, despite this huge potential there is no calibrated and validated model for estimating the effort required to develop a new mobile application. In this paper, we fill this gap by proposing a new parametric model for helping project managers to estimate the effort required to develop mobile applications. Improper software development effort estimation can result in project failures due to budget overruns and schedule slips. A reasonable effort estimate, on the other hand, can lead to a successful project by enabling solid planning and scheduling. The parametric effort estimation model proposed in this paper is calibrated using data of more than 160 real-life mobile applications developed by different software houses and freelancers. 20 cost drivers were initially identified and then forward stepwise regression was used to derive an effort estimation model. The obtained model uses 7 (out of the initial 20 cost drivers) and has an R2 value of 0.949. Results of validating this model using K-fold cross-validation (K=8) indicate that the average MMRE (using all 8 iterations) is 0.245 and the average PRED (30) is 77.75%. Results of comparison with COCOMO II indicate that our mobile-app-specific model provides more accurate estimates. On a validation dataset of 44 projects, the value of PRED(30) was 84% for our model as compared to only 4% for the general-purpose COCOMO II. Furthermore, the MMRE of our model is 0.20 as compared to 4.67 for COCOMO II.