Three Data Bipartitioning Methods to Improve Prediction of Time to Death

Accurate individual mortality risk prediction is a central challenge in survival analysis, especially in heterogeneous populations, where severe hazards can obscure subtler risks. We introduce a bipartition of survival data to improve prediction. Using data from the English Longitudinal Study of Ageing, we generate two relatively homogeneous subgroups—those at imminent risk and those at eventual risk of death—using three methods: disjunctive binary splits, an inverse oblique tree, and supervised principal components. For each method, a separate model is trained for each part, after which results are combined. Findings demonstrate improved prediction accuracy compared to overall models without partitioning.