It is hard for you to tell if the features you extracted are good or bad. ML algorithms will tell you.
Generally speaking, Deep Learning (CNN) will be the best one for you if you have enough number of images, where you do not need to conduct feature extraction - just feed them into the neural network.
Just in the case you do not have millions of images, you should think about "transfer learning".
If you want to stick to something traditional, you should keep using the algorithms available in Classification learner App.
But there is still room for you to work on: parameter tuning, PCA, feature selection... These can be done from the App! Do not worry, all you need is tick some boxes and wait.
