Feature subset selection for classification is inherently a bi-objective optimization problem, where the task is to obtain a feature subset yielding maximum possible area under the receiver operating characteristic curve by minimizing cardinality. In today’s world, an humungous amount of data is generated in all human activities. To mine such voluminous data, which is often high-dimensional, there is a need to develop parallel and scalable frameworks. In the first-of-its-kind study, we proposed and developed three chaotic bi-objective evolutionary algorithms based wrappers with a migration strategy under Spark, namely, (i) parallel chaotic non-dominated sorting algorithm (P-C-NSGA-II-IS), (ii) parallel chaotic non-dominated sorting particle swarm optimization (P-C-NSPSO-IS), and (iii) parallel chaotic multi-objective evolutionary algorithm based on decomposition (P-C-MOEA/D-IS). We employed logistic map and tent map for each of the parallel chaotic algorithm. The performance of the chaotic variants is compared with their corresponding parallel, non-chaotic algorithms. Throughout the study, AUC is computed by invoking the logistic regression classifier on various datasets. The experimental results demonstrate that P-C-NSGA-II-LM-IS, P-NSPSO-IS and P-NSGA-II-IS secured top-3 in terms of mean HV and Formula 1 racing based ranking. We also presented the statistical test of significance, empirical attainment plots, speedup analysis, and mean AUC obtained by the most repeated feature subset, and diversity analysis using hypervolume.