This experiment introduces a novel hybrid machine learning approach developed from improved Three Way Decisions and KNN algorithms in an environment with abrupt concept drift. Using crop yield datasets in simulated large-scale heterogeneous streaming datasets, the experiment compares the performance of the improved hybrid Three Way Decision and KNN algorithms with other existing abrupt concept drift generated datasets. The experiment was conducted under different parameters to determine the computational average accuracy on abrupt concept drift depending on the dynamic weights, instance runs on K Values, and computational accuracy of time. The Hybrid Three Way Decision_KNN (3WD_KNN) algorithms were tested against other state-of-the-art and existing algorithms to assess the accuracy performance, with results indicating steady and improved accuracy performance over other classification algorithms in the abrupt concept drift detection environments. The results show that the value of K has a significant influence on runtime, especially when considering the size of dynamic weights (W). The highest accuracy for the hybrid 3WD_KNN was achieved with an instance size of 300K and a dynamic weight of W = 300 at 95.63%, followed by 95.14% at an instance size of 100K with a dynamic weight of W = 500. Generally, as the K-values and dynamic weights (W) increase for the hybrid 3WD_KNN, it becomes evident that the computational runtime analysis increases significantly. However, the performance accuracy was noted to be efficient and more accurate than the rest of the tested algorithms.