Advancing SVM classification: Parallelizing conjugate gradient for monotonicity enforcement

•We propose a new PBCCG-RMC-SVM model considering prior monotonic domain knowledge.•Conjugate gradient is used to minimize subject to linear equality constraints.•The model is applicable to large-scale and complex problems.•The experiments show the proposed method outperforms the traditional SVM. Wi...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Knowledge-based systems Ročník 302; s. 112388
Hlavní autoři: Chuang, Hui-Chi, Chen, Chih-Chuan, Li, Sheng-Tun
Médium: Journal Article
Jazyk:angličtina
Vydáno: Elsevier B.V 25.10.2024
Témata:
Conjugate gradient algorithm
Monotonicity classification
Parallel strategy
Prior knowledge
Support vector machine
Prior knowledge
Parallel strategy
Support vector machine
Monotonicity classification
Conjugate gradient algorithm
ISSN:0950-7051
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:•We propose a new PBCCG-RMC-SVM model considering prior monotonic domain knowledge.•Conjugate gradient is used to minimize subject to linear equality constraints.•The model is applicable to large-scale and complex problems.•The experiments show the proposed method outperforms the traditional SVM. With the advent of multimedia, social media, and the Internet of Things, an unprecedented volume of data is being generated at a remarkable speed. Therefore, the application of data mining techniques has become essential for solving large-scale and increasingly complex problems. The integration of prior knowledge into data mining has also become a trending and challenging concern. This study proposed a novel support vector machine (SVM) model designed to address this concern. The model incorporates expert knowledge regarding the monotonic relations between response and predictor variables, represented through monotonicity constraints. In our approach, monotonic constraint SVMs were formulated by integrating regularization, monotonicity constraints, a box-constrained conjugate gradient, and a parallel strategy into a model to ensure solution uniqueness and boundedness. The model's ability to retain monotonicity was assessed using the frequency monotonicity rate. The experimental results highlight the feasibility and effectiveness of the proposed model, PBCCG-RMC-SVM, in addressing classification problems with monotonic prior knowledge. Additionally, the adoption of a parallel strategy accelerates the generation of analytical or prediction results, and therefore, the model can enable managers to make faster and more accurate decisions through data analysis.
ISSN:0950-7051
DOI:10.1016/j.knosys.2024.112388