K-means is a widely used clustering technique because of its straightforwardness. It aids in identifying unique, non-overlapping clusters. K-means is a significant unsupervised machine-learning approach used for clustering. One of the most difficult issues it faces is determining the optimal position of the first clusters’ centroids during the initialization phase. The selection of the first cluster centroid has a significant impact on its sensitivity. Choosing the appropriate value for the parameter K is a crucial concern. Choosing an adaptive optimal value for K poses a significant challenge, particularly in the context of increasing data volume, such as in big data problems and data streams. There are no effective and universally applicable methods for selecting the value of K in the existing literature. However, most of the relevant studies have chosen K randomly. This study presents novel methodology for determining the ideal value of K, taking into account the specific application and large-scale streaming data sets. The process involves constructing a statistical model by selecting an appropriate distribution to generate the observed data. The estimation of k is dependent on kernel density estimation, bandwidth estimation, and the probability density function of the data stream. Our approach shows an improvement between 4, 40% compared to Stream km++, IAPKM, IAPNN, DEN STREAM, DSTREAM, CLUSREE, DGSTREA.