Clustering algorithms are most probably and widely used analysis method for grouping agricultural data with high similarity. For example, one of the most widely used approaches in previous study is K-means, which is simpler, more versatile, and easier to understand and formulate. The only disadvantage of the K-means algorithm has always been that the predetermined set of cluster centres must be prepared ahead of time and provided as feedback. This paper addresses the issue of estimating cluster random centres for data segmentation and proposes a new method for locating appropriate random centres based on the frequency of attribute values. As a consequence of calculating cluster random centres, the number of iterations required to achieve optimum clusters in K-means will be reduced, as will the time required to shape the final clusters. The experimental findings show that our approach is efficient at estimating the right random cluster centres that indicate a fair separation of objects in the given database. The technique observation and comparative test results showed that the new strategy does not use present manual cluster centres, is more efficient in determining the original cluster centres, and therefore more successful in terms of time to converge the actual clusters especially in agricultural data bases.