Experimenting with K nearest neighbors
We attempt classification using K-nearest-neighbors (KNN) to increase the diversity of techniques used with the dataset.
Standardized data and a small number of neighbors
Having learned from logistic regression and neural networks, we only use standardized data for the KNN experiment. We also choose a small number of neighbors - 3 - to obtain results quickly. Finally, we leave the distance measure used by the KNN algorithm as the default (euclidean distance).
Classification accuracy
We find that the KNN classifier returns an accuracy greater than 99%. The accuracy is surprising given the default settings used, fast convergence of the classifier, and the small amount of effort expended in setting up the classifier.
Further investigation
We wonder if the KNN results are due to the specific type of processing applied to the data (1) or a lucky choice of parameters. KNN’s classification performance versus logistic regression and the neural network-based classifier motivate further investigation into the KNN’s training and classification characteristics.