Skip to main content

Table 3 Results of bootstrapped and cross-validated ascending variable introduction into a random survival forest classifier

From: Predicting general criminal recidivism in mentally disordered offenders using a random forest approach

Nested models OOB error sem IOE CVOE CVIOE
pcv 0.278 0.002 0.222 0.233 0.267
pcv + pco 0.276 0.002 0.003 0.255 −0.021
pcv + pco + age 0.221 0.002 0.054 0.23 0.025
pcv + pco + age + iofc 0.206 0.001 0.015 0.217 0.014
pcv + pco + age + iofc + icd1 0.204 0.002 0.002 0.211 0.005
pcv + pco + age + iofc + icd1 + div 0.197 0.002 0.007 0.193 0.019
  1. pcv: prior convictions; pco: prior criminal offences; iofc: index offence; icd1: ICD-10 F1; div: diversity of criminal history, sem: standard error of the mean, IOE: incremental OOB error, CVOE: cross-validated OOB error, CVIOE: cross-validated incremental OOB error.
  2. Note: the term incremental refers to the difference between two consecutive bootstrapped or cross-validated OOB classification errors.
  3. The bootstrapped OOB classification error was strictly monotonically decreasing (cf. column 2) as additional variables were introduced and as indicated by the constantly positive incremental OOB classification error (cf. column 4). An OOB classification error of 0.5 corresponds to tossing a coin. Except for prior criminal offences, each additional variable was also associated with a cross-validated OOB error reduction (cf. columns 5, 6).