In supervised learning, data classification is the method of categorizing data to facilitate data mining processes for informed decision-making. The central aim of a classification model is to accurately predict the categorical data for both familiar and unfamiliar instances. The classification models in machine learning are usually trained with datasets where instances are labeled. This paper explores an alternative way of constructing classification models based on the similarities of the instances rather than labels annotated by experts. The process of labeling data is a resource-intensive and time consuming process incredibly challenging when dealing with large datasets known as big data. In light of the proposed methodology clusters, the big data developing classifiers based on these clusters while bypassing the predefined class labels. This approach enhanced the performance of the classifier. Moreover, the generated clusters can be associated with the relevant class labels introducing a link between the unsupervised clustering and the supervised classification task. To validate our proposed approach, we gathered a diverse collection of data from Kaggle. For experimental analysis, we applied three widely recognized decision tree induction ID3 (Iterative Dichotomiser 3), C4.5 (extension of ID3 algorithm), CART (Classification & Regression Tree), NavieBayes classifier and Ensemble classifier(RandomForest, Bagging, Boosting). The outcomes of our investigation shed light on the potential of leveraging instance clustering for classification tasks, potentially revolutionizing the conventional paradigms of supervised learning in the domain of data mining and decision support.