Which machine learning is the best? The five top machine learning methods I prefer are dimension reduction, regression, ensembling methods, natural language processing and reinforcement learning.
Dimensions Reduction
Dimension reduction is important mainly because of its use in feature engineering and data pre-processing, where proper dimension reduction can greatly save time and space while keeping the most important information of the dataset. A good dataset that makes sense is just as important as having a powerful model, just as the saying goes – “rubbish in, rubbish out.”
Regression
Regression, being usually the most intuitive method to come right to mind, is both easy to interpret and fast to run. Though it might not give the highest accuracy, a simple linear regression gives a quick overview of the features in the dataset. Besides, more advanced regression models today, such as Bayesian regression models, prove beneficial in problem-solving.
Ensembling Methods
Ensembling methods are usually my next go-to after a simple regression model. Why? Because they usually give pretty high accuracy scores. With the help of tree models and gradient boosting, ensembling models can converge to a satisfactory final state. But, what is a satisfactory final state? What features are you looking for after running the ensembling method?
However, difficulty in interpretation and long running time make ensembling models unsuitable in some situations. Such as when understanding what key factors are important in the model is more important than giving an accurate prediction.
I pick NLP to be the 4th top ML method because of its distinctive application in data analytics nowadays.
What is NLP?
Natural language processing is a subfield of linguistics, computer science, and artificial intelligence concerned with the interactions between computers and human language, in particular how to program computers to process and analyze large amounts of natural language data.
Unlike the old days, more and more data we use today is unstructured and lots of data comes from news and talks. As information travels faster than ever. Markets today are greatly influenced by main trend opinions and public emotions, as well as policies. Being able to analyze those data becomes extremely essential in determining market trends.
Lastly, I picked reinforcement learning because of its self-evaluation characteristics. Learning from its own past actions and rewards, RL does not require many assumptions like regression, which is particularly useful in investing and trading in emerging and new markets. In conclusion, the trading bots trained using deep RL can interact with the market all day and optimize strategies based on different markets they are operating on.
