MNIST ("Modified National Institute of Standards and Technology") is the de facto “hello world” dataset of computer vision. Since its release in 1999, this classic dataset of handwritten images has served as the basis for benchmarking classification algorithms. A new machine learning techniques emerge, MNIST remains a reliable resource for researchers and learners alike.
In this competition, the goal is to correctly identify digits from a dataset of tens of thousands of handwritten images.
The competition is hosted on kaggle here.
This competition is evaluated on the categorization accuracy of your predictions (the percentage of images you get correct).
For this project, I have followed a down to the top approach wherein I have first applied very basic Machine learning algorithm without jumping to CNN. This would help one to understand why there arises a need of CNN & how it is able to easily achieve a good accuracy with image dataset.
I have uploaded my submission file in the submission folder.
Currently, I have achieved an accuracy of 99.671 & stands in top 5% of all kaggle submission
CODE
I have uploaded the code according to various approaches & have tried to explain it wherever necessary.
MNIST is an excellent dataset for anyone looking to enter or practice the basic concept of computer vision & deep learning.
Further, I will work so as to apply an ensemble of various models so that I can achieve a higher accuracy & will even try to use pre-trained models & transfer learning.