Approaches to Improving the Accuracy of Machine Learning Models in Requirements Elicitation Techniques Selection

TL;DR

This paper investigates how synthetic over-sampling techniques can enhance machine learning models' accuracy in selecting requirements elicitation methods, especially when training data is imbalanced, thereby aiding better planning in IT projects.

Contribution

It introduces the application of SMOTE to improve ML model performance in elicitation technique selection under imbalanced datasets, with experimental validation.

Findings

SMOTE improves ML model accuracy in imbalanced datasets

Proposed methods enhance feature importance selection

Effective for planning business analysis activities

Abstract

Selecting techniques is a crucial element of the business analysis approach planning in IT projects. Particular attention is paid to the choice of techniques for requirements elicitation. One of the promising methods for selecting techniques is using machine learning algorithms trained on the practitioners' experience considering different projects' contexts. The effectiveness of ML models is significantly affected by the balance of the training dataset, which is violated in the case of popular techniques. The paper aims to analyze the efficiency of the Synthetic Minority Over-sampling Technique usage in Machine Learning models for elicitation technique selection in case of the imbalanced training dataset and possible ways for positive feature importance selection. The computational experiment results confirmed the effectiveness of using the proposed approaches to improve the accuracy of machine learning models for selecting requirements elicitation techniques. Proposed approaches can be used to build Machine Learning models for business analysis activities planning in IT projects.