The cross-validation splits observations into K roughly equal-sized parts (for example, K = 4).
Starting from the root node containing all observations, each tree recursively splits observations into more homogeneous subsets.
For example, the Cox proportional hazards regression adopts a profile likelihood that cannot be split by observations.
Since the lamprey diverged before the teleost/tetrapodon split, these observations provide further support for this node as the root of the tree.
In practice, we can split the observations into two groups, assembled considering the estimates of the linear predictors and, for example, using their medians as cutpoints.
A decision tree generated from the GBM algorithm, as illustrated in Fig. 1, partitions the space of input variables by splitting the observations into homogenous quadrants and each tree split corresponds to an if-then rule for a predictor variable.
Therefore, each subject's person-years of observation were split into several observations by expanding data by 1-year age bands.
A decision tree was built by repeating best splits of observations of weather variables (e.g., rainfall, relative humidity, minimum temperature and maximum temperature), where a single splitting weather variable was selected at each split using the principle of impurity reduction.
To illustrate these differences, we split the observation window into three parts and analysed change in fatigue over these three periods: the pre-retirement period (year −1 v year −7), the transition to retirement (year +1 v year −1), and the post-retirement period (year +7 v year +1).
Therefore, we use a 2-fold method (e.g., [ 13, 14]), splitting all observations in the training cohort into two parts, one part for assigning gene weights, and another part for testing the regression parameter through a partial likelihood score test.
