DIABETES PREDICTION USING ANFIS WITH AUTOENCODING AND KALMAN FILTER
Restricted (Penn State Only)
- Author:
- Camero, Dabis
- Graduate Program:
- Electrical Engineering
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- June 17, 2024
- Committee Members:
- Aldo W Morales, Thesis Advisor/Co-Advisor
Truong Xuan Tran, Committee Member
Rafic Bachnak, Professor in Charge/Director of Graduate Studies
Jason Martis, Committee Member
Sannidhan M S, Committee Member - Keywords:
- Diabetes
predictive models
early diagnosis
World Health Organization (WHO)
Adaptive Neuro-Fuzzy Inference System (ANFIS)
Autoencoding
Kalman filtering
hybrid model
dimensionality reduction
feature extraction
BRFSS dataset
medical informatics
healthcare analytics
Explainable AI
chronic diseases
model generalizability
Computational Complexity
Outliers
Interpretability - Abstract:
- Diabetes is a chronic medical condition characterized by the body's inability to regulate blood sugar levels effectively due to issues with insulin production or function. According to the World Health Organization (WHO), nearly 422 million people worldwide are living with diabetes, with a substantial portion of cases remaining undiagnosed, particularly in middle and low income countries. The WHO Global Diabetes Compact aims to improve access to diagnosis and treatment to reduce diabetes risk and enhance care for those diagnosed. Given the increasing prevalence of diabetes and the associated health burden, there is a critical need for accurate predictive models to facilitate early diagnosis and intervention. The prevalence of diabetes has prompted significant research into predictive models that can accurately forecast the onset of this chronic condition. This thesis presents a unique approach to diabetes prediction by integrating an Adaptive Neuro-Fuzzy Inference System (ANFIS) model with Autoencoding and Kalman filtering techniques. The hybrid model aims to enhance prediction accuracy by leveraging the strengths of each component: the interpretability and adaptive capabilities of ANFIS, the dimensionality reduction and feature extraction process of autoencoders, and the noise reduction efficiency of Kalman filters. Initially, an autoencoder is employed to process and reduce the dimensionality of a high-dimensional diabetes dataset, ensuring that the most relevant features are retained for analysis. Subsequently, these features are fed into the ANFIS model, which combines the learning capabilities of neural networks with the fuzzy logic approach to handle the uncertainty and imprecision in the data. A Kalman filter is applied to refine the predictions further to mitigate the noise impact and enhance the data inputs' signal quality. The proposed methodology is validated through experimentation on the publicly available diabetes dataset produced by the Behavioral Risk Factor Surveillance System (BRFSS), comparing its performance against traditional prediction models. The results significantly improve prediction accuracy, robustness, and generalizability. This innovative framework not only contributes to the field of medical informatics but also offers a potential tool for early diagnosis and management of diabetes, ultimately aiming to improve patient outcomes and reduce healthcare costs. The findings of this research suggest that the integration of ANFIS, autoencoders, and Kalman filters represents a promising direction for future work in Predictive and Explainable healthcare analytics. Further exploration and optimization of this hybrid approach could lead to even more significant advancements in the predictive modeling of chronic diseases.