<i>K</i>-means partitioning approach to predict the error observations in small datasets
Pruthviraju Garikapati, K. Balamurugan, T.P. Latchoumi
Abstract
The partitioning algorithm was used to identify the uncertainty and the similarity in large sets of databases. K values are set based on the models. The effect of change in k values from the lowest to the highest level was analysed for a small set of databases that are acquired through machining AlSi7/63% SiC hybrid composite. An attempt has been made to identify the correlation between the k value clustered class and with a developed linear regression model. Further, the analysis was done to identify the critical machining observations that have a high error rate while on machining AlSi7/63% SiC hybrid composite using abrasive water jet at the varied parameters condition. Taguchi L27 orthogonal array observations are clustered into different groups with a k value of 2 to 8. The study was limited to k = 8 because at this level, clustered classes have very few observations that make unfit to predict the model.