During the Fall 2020 semester, I took a Data Mining class as an elective. I wanted to learn more about finding patterns within data sets too large to manage or analyze by hand. I learned about many new techniques and implemented the 6 steps of the Data Mining process with a scientific mindset. Towards the end of the course, we discussed the basics and variations of neural networks in machine learning. Knowing this has many interesting and important applications in modern careers, I decided to explore more of this area in my final project. After selecting the MNIST hand-written digits to practice identifying, I found MATLAB code online that had about 97% accuracy. Humans achieve about 99.8%. I improved the models accuracy to 98.12% by experimenting with its parameters. It only took about twice as long to train, which was good considering most processes were O(N). I documented the process and submitted a final paper with descriptive images of both concepts I learned and how my modified parameters affected the model. Because of this project, I better understand the complex processes and sub-goals of creating a neural network and hope to apply them on future neural networks with different fields and software languages.
Comments