İKİLİ YANITLI BOYLAMSAL VERİLERDE KLASİK VE MAKİNE ÖĞRENME MODELLERİNİN UYGULANMASI


Arş. Gör. RÜMEYSA RANA ARSLAN

Tez Türü: Yüksek Lisans

Tezin Yürütüldüğü Kurum: Orta Doğu Teknik Üniversitesi, Fen Edebiyat Fakültesi, İstatistik Bölümü, Türkiye

Tez Danışmanı: Özlem Ilk

Tezin Onay Tarihi: 2023

Tezin Dili: İngilizce

Özet:

Alzheimer’s disease (AD) is a significant global health issue that affects both individuals and society for older adults. The symptoms of the disease can be observed over time, making the structure longitudinal. Classical statistical models and machine learning algorithms can be used to analyze these datasets. This study consists of two parts: First, a real dataset is used to find the features affecting dementia status and compare the performances of models. Secondly, a simulation study based on the real dataset with a different number of subjects and an equal number of time points for each subject is conducted to apply and compare the model performances. The classical mixed models, their extended versions, and hybrid models, Boruta, GEE, GLMM, HGLM, GLMMLasso, GPBoost, GLMMTree, and HRF are used for both parts. As a result, GPBoost learns and classifies the dementia status well but overfits due to the small sample size in the dataset, and tree-based algorithms are efficient in predicting the dementia status when a new subject enters the study for the real dataset. For the simulation study, all methods have similar results, but HGLM, GPBoost and GLMMLasso algorithms have better performances regardless of the sample size and balance of the dataset.
Keywords: Longitudinal Data Analysis, Alzheimer’s Disease Prediction, Machine Learning Models, Mixed Effects Models