Hello, if you have any need, please feel free to consult us, this is my wechat: wx91due

EE5434 Homework 3

Submission format: your source codes (source code should be in a notebook file (.pynb) and also save your source code as a HTML file (.html).), and a simple but clearly written report (in pdf format). The submission site is still Canvas.

Data availability:

Please scroll down to the bottom of the above page to download the data. You can get both the extracted features (symmetry and intensity) and also the raw features. When we use “two features”, we refer to the extracted features. If we use “raw features”, we refer to the 256 grayscale values.

This homework contains the following components starting from linear classification models to more flexible classification. The features are from extracted features to raw features.

➔ 3-fold cross validation: divide your training data into three parts, use 2 parts for training and the third part for validation. Then apply the best one (based on validation accuracy) to the test set.2. (30 pts) Apply neural network and SVM for classification for all 10 digits, using the raw features as input. Describe your methods. Report and compare their classification accuracy using 3-fold cross-validation. Report the accuracy for each fold. Report the variance of the accuracy for three folds. Analyze the causal of the difference. The points are given based on your efforts of improving the performance.

3. (5 pts) Based on the above experiments, draw a conclusion about what is the best method for handwritten digit classification. Note that this conclusion must be supported by your analysis and experimental results. Otherwise, you won’t get full credit.