dsc478 final project

My project revolved around analyzing the 2018 stack overflow developer survey data. I had three main goals: first, to understand the groups in my data; second, to find frequent item pairs with various confidences and lifts; and lastly, to develop a decision tree regression which would predict your future salary based on your data.

Some post thoughts that I forgot to mention in the video, in addition to support, confidence, and lift I there was a 4th measurement statistic for frequent item pairs called conviction which was very interesting. When I noticed an item pair had a super high conviction, I realized I didn't need the data to confirm that for me, it was standard knowledge of things in the industry.

Also as far as hyper-tuning goes from my decision regression tree, I realized while filming the video I should have tested some scenarios when I transformed the target (aka y-value). I might have received better accuracies if I had taken the log() of the salary or some basic Gaussian normalization.

Скачать видео