We use accuracy because the previous chapter used accuracy. This allows us to have a apples-to-apples comparison. Additionally you may note that there is a lack of cross validation. That will be left as an exercise to the reader.
After training the neural network for two hours on a batch size of 50 and 150 epochs, I'm pleased to say I got a 99.87% accuracy. And this isn't even state of the art!
In the previous chapter, it took just 6.5 minutes to get a 97% accuracy. That additional 2% accuracy required a lot more time. This is a factor in real life. Often business decisions are a big factor in choosing ML algorithm.
