Temporal sampling forest ($\varvec{\textit{TS-F}}$): an ensemble temporal learner

Abstract

Ensemble learning is in favour of machine learning community due to its tolerance in handling divergence and biasness issues faced by a single learner. In this work, an ensemble temporal learner, namely temporal sampling forest (TS-F), is proposed. Building on the random forest, we consider its limitations in handling temporal classification tasks. Temporal data classification is an important area of machine learning and data mining, where it fills the gap of ordinary data classification when the observed datasets are temporally related across sequential and time domains. TS-F incorporated the temporal sampling (bagging) and temporal randomization procedures in the classical random forest, hence extending its ability to handle temporal data . TS-F was tested on 11 public sequential and temporal datasets from different domains . Experiments demonstrate that TS-F could provide promising results with average classification accuracy of 98 %, substantiating its ability to escalate the random forest performance in the application of temporal classification.