Survival Analysis Based on Fusion of Decisions from  Multiple Tree Structure: A Cutting-Edge Approach

Luz Marina Aguirre Paz^1,*, Jorge Viteri Moya¹, Rita Azucena Díaz Vásquez¹, Darvin Manuel Ramírez Guerra², Dekhkonov Burkhon³

¹Universidad Regional Autónoma de los Andes (UNIANDES), Ecuador

²Club  deportivo de la universidad San Martín de Porres, Peru

³TSUE Research department, Uzbekistan

Emails: direccionadmision@uniandes.edu.ec, direccionacademica@uniandes.edu.ec, ui.ritadiaz@uniandes.edu.ec u, dramirezg1978@gmail.com; b.dekhkonov@tsue.uz

Abstract

Survival analysis remains an important area in predictive modeling, especially in cases where event timing information is critical.  This work presents a research effort to investigate the application of LightGBM, a high-performance high-throughput model, to conduct an improved fusion of decisions from multiple trees to reach survival analysis. Our objective is to address the challenge of developing correct predictive models while advancing computational effectiveness.  Based on a case study of live disaster scenarios, the proposed approach applies and compares LightGBM with traditional prediction methods, which involve careful design engineering, and model training with LightGBM tree structure refinement. The results obtained from fair experimentation and comprehensive predictive performance evaluation demonstrate the robustness of LightGBM in increasing the accuracy of relevant classification tasks toward survival analysis. Furthermore, the findings highlighted that the combination of excellent tree depth for cutting and multi-thread optimization promotes efficient computational complexity and prediction accuracy.

Keywords: Survival analysis; Information Fusion; Predictive modeling; Machine learning; Prognostic analysis; Time-to-event analysis; Novel prognostic models.