Granular Ball Twin Support Vector Machine With Pinball Loss Function

Abstract

Alzheimer&#x2019

s disease (AD) and Schizophrenia (SCZ) are prominent neurodegenerative conditions and leading causes of dementia, resulting in progressive cognitive decline and memory loss. Several studies reveal that early detection and intervention can slow the progression of AD and SCZ. Numerous machine learning algorithms including twin support vector machine (TSVM) have been proposed for the early diagnosis of AD and SCZ. However, TSVM grapples with significant challenges: 1) TSVM relies on the hinge loss function, resulting in susceptibility to noise and instability

2) TSVM encounters challenges in effectively handling large datasets, attributed to its computational complexity and dependence on matrix inversions. Keeping in view the aforementioned challenges, in this article, we propose a novel granular ball twin support vector machine with pinball loss function (Pin-GBTSVM). Pin-GBTSVM employs granular balls, as opposed to individual data points, as inputs for constructing a classifier, while also leveraging the pinball loss function to attain a heightened level of noise insensitivity. The proposed Pin-GBTSVM persists in facing challenges associated with the absence of integration of the structural risk minimization (SRM) principle and the requirement for matrix inversions. We further propose a novel large-scale Pin-GBTSVM (Pin- LGBTSVM). Pin-LGBTSVM achieves two crucial objectives: 1) it eliminates the necessity for matrix inversions, streamlining the computational efficiency of Pin-GBTSVM

and 2) it integrates the SRM principle by incorporating regularization terms, effectively addressing the concern of overfitting. Experiments are conducted on University of California Irvine (UCI), knowledge extraction based on evolutionary learning (KEEL), and normally distributed clustered (NDC) benchmark datasets, where both the proposed Pin-GBTSVM and Pin-LGBTSVM models consistently outperform the baseline models in terms of generalization performance. Furthermore, we implemented the proposed Pin-GBTSVM and Pin-LGBTSVM models on SCZ and Alzheimer&#x2019

s Disease Neuroimaging Initiative (ADNI) datasets, showcasing the model&#x2019

s efficacy in real-world applications. IEEE