Kaggle submission of the Digit Recognizer challenge using multiple ML Algorithms and their result comparisons.
My first submission was using K-Nearest Neighbour Algorithm (taking K = 3), which gave an accuracy of 96.857%. The accuracy went up to 96.943% when applying PCA before KNN.
The next submission was made by using PCA with SVM, which gave an accuracy of 98.057%, much better than the previous predictions.
Next prediction was made using decision tree, which brought the accuracy down to 85.286%.
The Training and Testing Dataset can be downloaded from Kaggle itself:
https://www.kaggle.com/c/digit-recognizer
- Scikit-Learn
- Pandas
- Numpy