Boosting Methods for Interval-censored Data with Regression and Classification
Authors: Yuan Bian, Grace Yi, Wenqing He
ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We validate our methods through extensive experiments on both synthetic and real-world datasets. Results show that L2Boost-CUT and L2Boost-IMP offer robust and scalable solutions for handling interval-censored data and enhancing the generalizability of boosting algorithms. Empirical studies demonstrate robust performance across various finite-sample scenarios, highlighting the practical utility of our approaches. |
| Researcher Affiliation | Academia | Yuan Bian1, Grace Y. Yi1,2,3 , Wenqing He1 1Department of Statistical and Actuarial Sciences, University of Western Ontario 2Department of Computer Science, University of Western Ontario 3Vector Institute, Canada EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1 L2Boost-CUT 1: Take f (0) = arg minh n 1 Pn i=1 n ˆY1(Oi) h(Xi) o2 and set η = n w for a given w 1; 2: for iteration t with t = 1, 2, . . . do 3: (i) calculate ˆL Oi, f (t 1)(Xi) ˆL(u,v) v u=Oi,v=f (t 1)(Xi) for i = 1, . . . , n; 4: (ii) find ˆh(t) = arg minh(t) n 1 Pn i=1 n ˆL Oi, f (t 1)(Xi) h(t)(Xi) o2 ; 5: (iii) for regression tasks, update f (t)(Xi) as (3) for i = 1, . . . , n; 6: for classification tasks, update f (t)(Xi) as (5) for i = 1, . . . , n; 7: if at iteration t, n 1 n X i=1 ˆL Oi, f ( t)(Xi) n 1 n X i=1 ˆL Oi, f ( t 1)(Xi) η (12) 8: then stop iteration and define the final estimator as ˆf CUT n = f ( t 1). 9: end if 10: end for |
| Open Source Code | Yes | Algorithm 1 outlines a pseudo-code for obtaining ˆf CUT n ; its implementation code is available on Git Hub at https://github.com/krisyuanbian/L2BOOST-IC. |
| Open Datasets | Yes | We apply the proposed CUT and IMP methods as well as the N method to analyze two datasets, Signal Tandmobiel data and Bangkok HIV data, whose details are included in Appendix F.4. ...Signal Tandmobiel project a longitudinal intervention health promotion study in Flanders (Belgium): Baseline and first year results. European Journal of Paediatric Dentistry, 2:87 96, 2000. ...The study enrolled 1124 participants who were HIV negative at the time of enrollment. These participants were repeatedly tested for HIV at approximately four-month intervals over the study period. The response variable Yi represents the time when participant i was first tested positive for HIV. Among those participants, 991 were right-censored, meaning they were never tested positive during the study period, while the remaining were interval-censored, meaning the exact time of seroconversion is only known to occur between two testing intervals. The features are defined as follows: Xi1 = 0 if the participant is a female and 1 otherwise; Xi2 = 0 if the participant has a history of injecting drug use and 1 otherwise; and Xi3 represents the scaled age at enrollment. Vanichseni et al., 2001 |
| Dataset Splits | Yes | Synthetic data are split into training and test datasets in a 4 : 1 ratio. We assess the performance of each method using sample-based maximum absolute error (SMax AE), sample-based mean squared error (SMSq E), and sample-based Kendall s τ (SKDT), for regression tasks, along with sensitivity and specificity for classification tasks. The dataset is divided into OTR {{Yi, Xi, ϕ(Xi), ui,j} : i = 1, . . . , n1; j = 1, . . . , m} and OTE {{Yi, Xi, ϕ(Xi), ui,j} : i = n1 + 1, . . . , n1 + n2; j = 1, . . . , m} in a 4 : 1 ratio, where n1 = 400 and n2 = 100. |
| Hardware Specification | No | Computing times (in second) are reported in Table F.1 for three sample sizes, where we separately display computing time for implementing ICRF from that for implementing both unbiased transformation and boosting (UT + B). The implementation of the proposed methods requires a lot longer time than that for the O, R, and N methods, as expected. Table F.1: Computing times (in second) using a cluster with 1 node, where UT and B represent the procedure corresponding to unbiased transformation and boosting, respectively. |
| Software Dependencies | No | No specific software versions are mentioned. The paper refers to R packages 'survival' and 'ICcforest' without version numbers, and describes internal parameters for ICRF and kernel functions but does not specify external software dependencies with versions. |
| Experiment Setup | Yes | Experimental setup. Each experimental setup involves conducting 300 experiments with a sample size n. For i = 1, . . . , n, let Xi = (X1,i, . . . , Xp,i) , where the Xl,i are independently drawn from the uniform distribution over [0, 1] for l = 1, . . . , p and i = 1, . . . , n. The responses Yi are then independently generated from an accelerated failure time (AFT) model (Sun, 2006), given by (15), where g(u) = log u, and the error terms ϵi are independently generated from either a normal distribution N(0, σ2) with variance σ2 or the logistic distribution with location and scale parameters set as 0 and 1/8, respectively. For i = 1, . . . , n, we generate m monitoring times independently from a uniform distribution over [0, τ], and then order them as ui,1 < ui,2 < . . . < ui,m. We set n = 500, σ = 0.25, p = 1, τ = 6, m = 3, and ϕ(Xi) = β0|Xi 0.5| + β1X3 i + β2 sin(πXi), with β0 = 1, β1 = 0.8, and β2 = 0.8. Regardless of the value of n, we set w = 5 for Algorithm 1 as a stopping criterion, and take cubic smoothing splines as base learners with r = v = 2 in Section 3.3. ... for (B.1), we set df = 20, and replace Ψ in (14) with Ψu and u = 0.01. ... For ICRF, we specify the splitting rule as GWRS, ... take K = 5 and D = 300. ... nmin is the minimum size of terminal nodes, set to 6. |