Analogizing Time Complexity of KNN and CNN in Recognizing Handwritten Digits

Handwritten identification of digit, analysis of pattern has been a major area of research in the field of character recognition. Several models are formed by changing various values of weights when applied through the neural network. The paper analyzes the study of time complexity in two different algorithms KNN and CNN. The K-Nearest Neighbor Algorithm is used as a classifier capable of computing the Euclidean distance between data set input images. The dataset is fetched for training through neural network contains various (28 x 28) pixel size images and therefore, our first layer of neural network contains 784 neurons as input. We will analyze these images by varying values so as to obtain output layer of our network 10 neurons, each neuron if fixed gives any output between 0 to 9. After reading in data appropriately from MNIST and testing it on Gaussian distribution, KNN classifier then presented the result with Python tool. So, to avoid such a long waiting another algorithm Convolutional Neural Networks has been used. CNN has been implemented on Keras including Tensorflow and produces accuracy. It is then shown that KNN and CNN perform competitively with their respective algorithm on this dataset, while CNN produces high accuracy than KNN and hence chosen as a better approach.