This project was developed for the Machine Learning course. Study only!
The train.csv and test.csv files contain images of the MNIST dataset, in grayscale, of the digits 0, 1, 4 and 5 written by hand. Each image is made up of 28 rows and 28 columns in one total of 784 pixels. Each pixel has a unique associated value, which indicates its shade of gray. The higher this value is, the darker the pixel. The values of each pixel are in the closed interval [0, 255].
Machine Learning algorithms used:
- Perceptron
- Linear Regression
- Logistic Regression
- Regularizated Logistic Regression
These algorithms were also developed
Steps:
1 - Sample Size Reduction using vertical symmetry, horizontal symmetry and intensity
2 - Multi-class Classification (One vs All)
3 - Weight Decay Heuristic