AI-Powered Software Quality Assessment and Defect Prediction for Healthcare IoT Devices
DOI:
https://doi.org/10.65477/ijmdas.2025.v1.i3.02Keywords:
Software defect prediction, Machine learning, Ensemble learning, XGBoost, Explainable artificial intelligence, Software quality estimation.”Abstract
It is very important for software defect prediction to improve software quality, reduce cost of maintenance and identify the modules which are prone to software defect in software development process. Traditional classification techniques normally confront the problems of modelling complicated interactions between software metrics and detecting malfunctioning modules correctly when dealing with imbalanced information. The approach uses two benchmark datasets, KC1 and PC1, which are available from the NASA PROMISE repository, and contain software quality attributes, such as code size, cyclomatic complexity, Halstead metrics and control-flow characteristics. The preprocessing pipeline consists of data cleansing, missing value handling, duplication removal, feature scaling, and class imbalance control to boost the reliability of classification. The ML methods included are Bagging, RF, MLP, Radial Basis Function, NB, MNB, SVM, XGBoost and CatBoost. LIME and SHAP are used to make predictions that are interpretable and FastAPI is used to deploy the model. The accuracy, precision, recall, F1-score, ROC-AUC, PR-AUC and Matthews Correlation Coefficient are used to measure performance. According to provisional estimations, CatBoost has the highest accuracy (89.7 % on KC1 and 95.8 % on PC1) which is consistent with what was expected of it. The proposed method combines the complex ensemble classification, explainable AI and API deployment that facilitates robust software fault estimation.
Downloads
Published
Issue
Section
License
Copyright (c) 2025 Author(s)

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
