An Overview Of My Wine AI

I decided to make an AI that predicts the quality of red wine recently. The first step I took was to remove unnecessary features, which I did by using an ANOVA test. I found that residual sugars didn't have an impact on wine quality, so I removed it to make my data less sparse. After that, I removed the wines with a quality of 3, 4, and 8. The rationale was that there should be 10 observations per feature, which means a minimum of 100 observations a label. There were less than 100 wines with a quality of 3, 4, and 8 combined, so they weren't usable. After that, I used a Knn algorithm to classify them. It took a bit of playing around with the code, but I was able to get an AI that recorded an accuracy of 72.5%. Based on looking through the code of other people who used the same dataset, it seems that this is something of an upper limit for accuracy with this dataset.