What you get:
- 10 ideas for machine learning projects.
- Free datasets to use for your projects.
- Examples of libraries and algorithms for each case.
If you are here with the hope that I will show you a method to get rich by predicting stock prices, sorry, I'm don't know the solution. Even if I knew, I would never tell you!
Even if you think it is impossible to predict the future, it is not. In fact, that is one of the reasons we invented math. Probably you have heard about something called differential equations (DE). If you are an engineer, you probably had to pass a painful exam of DE (I did). DE are just one way to model the temporal behavior of a system.
Another method is called Recurrent Neural Networks (RNNs), which is a deep learning model designed to address sequential data problems. A variant of RNNs is the Long-Short Term Memory (LSTM) which is specially suited for time-series prediction problems. It is internally composed by two kind of units, hidden units, which as in CNNs contain a hidden representation of the inputs, and gated units, which control the amount of information that flows from the inputs, to the hidden units and to the outputs.
The gated structure of LSTMs, make them very efficient when working with large time series. They are able to understand long-term dependencies in temporal data, controlling which information should be "remembered" and which "forgotten".
I created a simple tutorial using Keras with Tensorflow back-end to predict the stock prices of Microsoft from 2012 to 2017. I used an LSTM and two variants of the same family: Bi-directional LSTM and GRU.
In the figure, the model is able to track the true value (in green) with the prediction in the test set (in blue), given a training set of past values (in red). This code is probably not enough to earn money in the stock market, but it will give you a good baseline if you want to enter the exciting world of financial services with AI.