Choice of Machine Learning Classifier depends upon the data set. When a substantial amount of training data is available, boosting is generally effective. Random trees are frequently quite effective, and they can also be used to do regression.
K-nearest neighbors - the most basic thing you can perform, but can be slow and demands a lot of memory.
Slow to train but incredibly fast to operate, neural networks are still the best performer for letter recognition.
With minimal data, SVM is one of the best
With categorical/binomial data, Bayesian works best.
Complex non-linear classification can be handled using neural nets and SVMs.
Thus, there is no benchmark or pre-defined solutions, it is all about performing iterations over the model, improving the results with hyperparameter tuning and then finalizing the model after continued improvements and iterations.