Ekpo, Raphael Henshaw and Covenant University, Theses (2024) MAP REDUCE SECURITY MODEL FOR ASTHMA PREDICTION IN CHILDREN USING FEDERATED XGBOOST. ["eprint_fieldopt_thesis_type_phd" not defined] thesis, Covenant University.
![[img]](http://eprints.covenantuniversity.edu.ng/style/images/fileicons/application_pdf.png) |
PDF
Download (319kB) |
Abstract
A substantial number of asthma development prediction models in children, such as conventional methods involving risk factors, logistic regression, the hybrid of statistical methods, and machine learning based approaches, exist. The problem associated with conventional methods of asthma prediction in children is low predictive accuracy of the model. However, using centralised machine learning approaches in healthcare requires training the learning models on large datasets. Besides cost, data privacy and security represent the main problems with centralised machine learning. The objective of this study is to develop a Map Reduce Security for asthma prediction in children using a federated XGBoost as a response to the aforementioned limitations associated with the existing asthma prediction model for children. This study leveraged two diverse datasets: the Nigerian Hospital Asthma dataset and the National Survey of Children's Health dataset for benchmarking purposes. After preprocessing the dataset, the symmetrical uncertainty and normalization interaction gain, as well as the undersampling approach, were employed for feature selection and dataset balancing. The system was trained, and tested using Federated Artificial Intelligence (AI) Technology Enabler (FATE) and extended to XGBoost model with one central server for federated algorithm averaging. The map reduce security measure was employed for the input data during training to avoid data leakage. The study was implemented with the Python programming language on Google Collaboratory (Collab) environment. The results of the analysis showed considerable high accuracy of 0.98, precision (0.98%), recall (0.98%), and F1-score (0.99%) for the asthmatic class and precision of 0.98%, as well as F1-score of 0.99% for the non-asthmatic class. The implemented model was benchmarked with the existing study on asthma prediction model in children using the same NSCH dataset of 23 features and 50212 samples. This study achieved a prediction accuracy of 93.8%, outperforming the existing models. The simulated attack on the implemented model results show that the model correctly identified whether a data point was used in the training process with predictive accuracy of 97.94%, membership inference attack, and data loss of 0.014%, respectively. In conclusion, the study finds that the model developed by the researchers performed better in predicting childhood asthma than some of the state-of-the-art machine learning algorithms.
| Item Type: | Thesis (["eprint_fieldopt_thesis_type_phd" not defined]) |
|---|---|
| Uncontrolled Keywords: | Asthma prediction model, Asthma in Children, Encryption, Federated Learning, Performance Metrics and Machine Learning. |
| Subjects: | Q Science > QA Mathematics Q Science > QA Mathematics > QA75 Electronic computers. Computer science |
| Divisions: | Faculty of Engineering, Science and Mathematics > School of Electronics and Computer Science |
| Depositing User: | Patricia Nwokealisi |
| Date Deposited: | 21 Jun 2024 12:57 |
| Last Modified: | 21 Jun 2024 12:57 |
| URI: | http://eprints.covenantuniversity.edu.ng/id/eprint/18115 |
Actions (login required)
 |
View Item |
