An Improved K-NN Algorithm Through Class Discernibility and Cohesiveness

The K-Nearest Neighbor (K-NN) is a primarily chosen method when it comes to the object classification, disease interpretation, and various other fields. In numerous cases, K-NN classifier uses the only parameter as K value, which is the number of nearest neighbors to decide the class of the instance and this appears to be insufficient. Within this study, we have looked at the initial K-Nearest Neighbor algorithm and also proposed modified K-NN algorithm to identify various ailments. Enhancing precision of the initial K-Nearest Neighbor algorithm, this specific suggested method consists of instance weights as an added parameter to determine the class of the example. This study presented a novel technique to assign weights, which utilizes the information from the structure of the data set and assigns weights to every instance relying on the priority of the instance in class discernibility. In this approach, we have included an additional metric "average density" together with "discernibility" to calculate an index which is used as a measure also with the value of K. The practice results obtained from UCI repository reveals that this classifier carries out much better than the traditional K-NN and preserve steady accuracy.