Software Enhancement Effort Prediction Using Machine-Learning Techniques: A Systematic Mapping Study

Abstract

Accurate prediction of software enhancement effort is a key success in software project management. To increase the accuracy of estimates, several proposals used machine-learning (ML) techniques for predicting the software project effort. However, there is no clear evidence for determining which techniques to select for predicting more accurate effort within the context of enhancement projects. This paper aims to present a systematic mapping study (SMS) related to the use of ML techniques for predicting software enhancement effort (SEME). A SMS was performed by reviewing relevant papers from 1995 through 2020. We followed well-known guidelines. We selected 30 relevant studies; 19 from journals and 11 conferences proceedings through 4 search engines. Some of the key findings indicate that (1) there is relatively little activity in the area of SEME, (2) most of the successful studies cited focused on regression problems for enhancement maintenance effort prediction, (3) SEME is the dependent variable the most commonly used in software enhancement project planning, and the enhancement size (or the functional change size) is the most used independent variables, (4) several private datasets were used in the selected studies, and there is a growing demand for the use of commonly published datasets, and (5) only single models were employed for SEME prediction. Results indicate that much more work is needed to develop repositories in all prediction models. Based on the findings obtained in this SMS, estimators should be aware that SEME using ML techniques as part of non-algorithmic models demonstrated increased accuracy prediction over the algorithmic models. The use of ML techniques generally provides a reasonable accuracy when using the enhancement functional size as independent variables.

AI-Assisted Fatigue and Stamina Control for Performance Sports on IMU-Generated Multivariate Times Series Datasets

“…In algorithms, we can easily integrate a deep neural network (DNN) [ 17 , 18 ] that includes convolutional layers for spatial feature extraction and recurrent layers, like LSTM [ 19 , 20 ] or GRU, for temporal dependencies. The ensemble method for stamina and fatigue consists of various ML models such as gradient boosting, bagging, or additional neural networks (NNs) [ 21 ], each trained to predict stamina and fatigue from the high-level features learned by the deep learning model. The ensemble methods not only capture different aspects of the complex physiological data but also help to improve the robustness and accuracy of the predictions for stamina and fatigue.…”