This article presents, an adaptive improved binary particle swarm optimisation-based learnable Bayesian classifier for dimensionally reduced microarray data. In the first fold of this two-folded work, the problem of dimension has been reduced by unsupervised method of feature reduction. The k-means clustering algorithm has been applied on the microarray data to group functionally redundant genes followed by application of signal-to-noise-ratio ranking technique to generate an intermediate feature subset consisting of most relevant and non-redundant feature subsets. In the second fold, the feature subset has been given to an adaptive binary particle swarm optimisation-based learnable Bayesian classifier for simultaneous selection of features and classification. We have conducted an extensive experimental work on a few benchmark datasets to validate its classification accuracy with and without reducing the dimensionality of microarray data. It was observed that our method is not only accepted as a good classifier over methods, which are considered here for comparison but also be treated as an alternative method of reducing dimension of the problem.