A Comparative Statistical Analysis Between ML Algorithms & DNN Techniques Using MNIST Dataset

Handwritten digit recognition has been around since a very long time and it has always been a topic of attention among the students and researchers. This job of recognizing handwritten content has great importance as it helps in recognizing various type of important information which are filled by hand but are not clearly understood. While solving this problem, the most common challenge is that the digits written by hands are of different size, orientation or thickness. Inspite of all these challenges, various Machine Learning Algorithms & Deep Neural Network Techniques are trying harder to achieve higher accuracy. In this report, we have tried to compare the accuracies of most common and popular ML Algorithms which includes SVM (Support Vector Machine), Random Forest Classifier (RFC) & KNN (K-Nearest Neighbour) with DNN Techniques (Deep Neural Network Technique) also termed as CNN (Convolutional Neural Network) using Keras. We have applied these algorithms and techniques on MNIST dataset. With the help of the results obtained, we have plotted some graphs to show differences between ML (Machine Learning Algorithms) & DNN (Deep Neural Network Techniques). In this research work, we got the accuracy of approx 97.86% using SVM, 97.10% using KNN, 97.05% using RFC while on the other hand, by using CNN, we got 99.98% accuracy on validation image data. Also, we were able to achieve accuracy approx , 96.85% using RFC, 97.73% using SVM, 96.80% using KNN while on the other hand, by using CNN, we got 98.72% accuracy on test image data.