Project

• Designed and explained a Convolution Neural Network (CNN) in R keras
• Used graphics to visualize the concept of CNN
• The final report is available here

(Further model improvements)

• Ensemble CNNs with censorship to boost model accuracy

Model improvement

Ensemble

1. Applied 4 censors seperately to the images
2. Trained a CNN for each censorship
3. Ensembled them and got 91.19% accuracy (0.14% improvement on the base model)

Source of dataset

The fashion mnist dataset is obtainable at Zalando Research. It is a dataset with 60000 (train) + 10000 (validate) = 70000 (total) Zalando's article images.

R version

R version 4.0.4

Machine used

Macbook Air (13-inch, Early 2015)

Replicate the results

Step 1:	Install R and RStudio
Step 2:	Download R code file R_combine.R
Step 3:	Run it and install any necessary libraries
(Note: Allow >12 hours runtime for any computers without Nvidia graphics card)
Step 4:	Download R markdown file Report.Rmd
(Note: R code should be run to produce a RData file before producing the R markdown report)
Step 5:	Run it and install any necessary libraries

Lastly...

The project content is totally original and solely for academic purposes.
This project is a great experience and fun to work on, so do not plagiarize.