Swarm intelligence-based model for improving prediction performance of low-expectation teams in educational software...

Area of Science:

• Software Engineering Education
• Machine Learning Applications
• Team Performance Prediction

Background:

• Software engineering projects are crucial for practical skill development in students.
• Low-expectation teams often face challenges, necessitating early performance prediction for educational success.
• Existing methods for early performance prediction in software engineering teams have limitations.

Purpose of the Study:

• To develop and evaluate a swarm intelligence-based model for early prediction of low-expectation software engineering team performance.
• To improve prediction accuracy by identifying a reduced set of relevant software product and process features.
• To provide instructors with tools to proactively address team challenges and prevent project failure.

Main Methods:

• Implementation of a Particle Swarm Optimization-K Nearest Neighbours (PSO-KNN) model.
• Feature selection to identify less than 40 relevant software product and process attributes.
• Experimental validation using the Software Engineering Team Assessment and Prediction (SETAP) dataset.

Main Results:

• The proposed PSO-KNN model demonstrated superior prediction performance compared to traditional Machine Learning classifiers.
• The model successfully reduced the number of features while maintaining or improving prediction accuracy.
• Early prediction of team performance was achieved in the initial phases of the Software Development Life Cycle (SDLC).

Conclusions:

• The PSO-KNN model offers a promising approach for enhancing software engineering education through accurate and early team performance prediction.
• Reducing feature dimensionality can lead to more efficient and effective predictive models.
• Proactive intervention based on early predictions can significantly improve software project outcomes for low-expectation teams.