The advantages of KNN algorithm are:

* KNN algorithm is very simple to implement. By using appropriate nearest neighbor search algorithm makes k-NN computationally tractable even for larger data sets, dealing with as high as 60000 data sets in our case.

* KNN algorithm is analytically tractable.

* KNN algorithm is highly adaptive to local information. We have used Euclidean distance metrics in our KNN algorithm to estimate the closest data points. Thus the algorithm is able to take full advantage of the local information in the training datasets to form a highly nonlinear, highly adaptive decision boundaries for each data point.

* KNN algorithm is easily implemented in parallel. For each data set to be evaluated, it checks against the training data-set for the k nearest neighbors around the data set to be evaluated. Since each of these data set is independent of the others, the search and validation can be implemented in parallel.

* Choosing higher value of K will yield smoother decision regions and provide probabilistic information.

The disadvantages of KNN algorithm are:

* KNN algorithm is computationally very intensive as it needs to search through training datasets (of size 60000) to find the nearest neighbors and determine the closest ones for each of the testing data sets which needs to be classified.

* Since the datasets is huge (as we are dealing with 60000 training sets), storage will be a concern as we need to store all the data sets in training sample and use when we need to validate the test datasets.

* As dimensionality of k increases, it becomes difficult to calculate the ‘Euclidean’ distance and finding the appropriate space to find the relevant datasets for classification and computational cost also increases.

* KNN algorithm is considered as a lazy learner as it does not learn anything from the training data and simply uses the training data for classification.

* KNN algorithm will compute the distance and sort all the training data at each prediction, which becomes slow in our case as we are dealing with large training dataset of 60000 samples.

* Since the algorithm does not learn anything from the training data, the KNN classifier may not be robust to noisy data. By changing the value of K, the resulting predicted class label can change as well.