Multi-objective genetic-clustering algorithms are based on optimization which optimizes several objectives simultaneously. In multi-objective optimization problem (MOP), different objective function may have different properties. In the previous paper, multi-objective optimization on neighbourhood learning using k-means genetic algorithm (NLMOGA), was proposed and applied to several real-life data sets. This research paper aims to extend NLMOGA by maximizing the compactness and the accuracy of the solution through constraint feature selection on the selected sub-population. A new population is generated using NLMOGA and a constrained feature selection is applied to each sub-population. This method is developed to determine a suitable or closest set of objects from the group objects which will improve the robustness of NLMOGA for different instances of MOPs. In NLMOGA, a solution is selected from global population repository and then neighbourhood learning is made to promote the evolution of each objective for the selected solution. The effectiveness of this approach is evaluated with various real-life benchmark gene expression data sets.