Guided Project: Building a Handwritten Digits Classifier - K-NN, Neural Networks, Decision Trees, and Random Forests!

I spent a very long time on this complex project. I’m honestly not 100% sure what to make of the results, and I tried my best to visualize the weights, but I believe all of the models tested were severely overfitting the data, thus leading to the inaccurate data at the end. But I also believe that was partially the point of the project, to see why image classification is so difficult.

Note: I have a sensitivity to bright light, so I changed my jupyterthemes to use a darker theme, so if some of my graphs and visualizations looked weird, that’s why.

Anyway, literally any feedback would be great! I probably over-engineered/over-complicated some functions, but it all worked out in the end. Any feedback on how to make my visualizations better would be nice!


Handwritten Digits Classifier.ipynb (1.2 MB)

Click here to view the jupyter notebook file in a new tab


Hi @noah.gampe

Thanks for sharing your project with DQ community. The project looks great and well styled. Personally , haven’t done this project yet but with some experience I have in machine learning, I can affirm that this project is just on another level, the visualization process is very admirable , the explanations given , the findings are very informing. I am confident that this project will serve as guideline to most of the students in this community. Very soon I will be doing the project to give some suggestions if there will be any. Keep it up mate for the good.

1 Like

Hi @noah.gampe:

Great observations. I like how you compared the 4 methods at the end of the project and explained some of the advantages and disadvantages of various methods. Perhaps you could also evaluate the applicability of these after reviewing the results of each model type/method(KNN, Neural Net etc.), as not all of these may apply to that particular model given the results (when compared with the results of the other 3).

Overall, good work with keeping the code modular and readable with the comments!

1 Like

Thank you very much for your kind words! This definitely makes me happy to know that people are looking at it and finding it helpful! :smiley:

1 Like

Cheers! Thanks for the kindness! I always struggle with remembering the accuracies/scores of previous tests, so I wanted to have a comparison of all of them side by side to make it easier for me to remember, haha! This project was really difficult for me to conceptualize in my head, and honestly I did have to look at a lot of other people’s examples to make sure I was on the right path. The visualization at the very end (visualizing the weights) was perhaps the most confusing part for me. I hope I can spend more time on this in the future, because I do think it’s a fascinating project. :slight_smile:

1 Like

@ noah.gampe Very well worked project. Gives clear insights of KNN and K-Fold models.

1 Like

Cheers! I appreciate it! I’m about to post my next project, so I hope it’s just as helpful! :smiley:

1 Like

@noah.gampe Look forward to it :+1:

1 Like