A suitable method to predict next binary outcome based on the patterns of past binary data

Question

I have a series of binary data ( W and L values which are trade results of a trading platform) and I need to be able to predict the next value (Whether W or L) based on the past patterns. What would be the best method to accomplish this in Matlab o python.

I have already tried a basic pattern matching algorithm developed by my self. What I do there is get an input sequence of 5 outcomes and match it with all past data to get a probability of the 6th outcome. However the accuracy of that method is close to 30% which is not suitable for my prediction. That is a very basic method, I'm sure there must be other machine learning methods which would give more accurate results.

Basically What I need is, I have a past data sequence [ W, L, W , W , L , W ......up to 4300 points ] like this. And my system generates new data feeds like this [ W , L, L ,W ...] what I need is to predict the value of the next data, by matching the patterns of my current data feed to the past 4300 data points.

Answer 1

You can try using Markov Chains (I suggest you to start here):

• https://www.datacamp.com/community/tutorials/markov-chains-python-tutorial
• https://towardsdatascience.com/how-to-build-a-market-simulator-using-markov-chains-and-python-7923256f8d29
• https://towardsdatascience.com/when-to-buy-the-dip-e2e128d737a7

Or you can try another approach training a neural network, and then using it to predict (i.e. using LSTM):

• https://medium.com/@kushal.sharma/lstm-network-using-keras-for-sequence-prediction-550b5bebae2c
• https://hackernoon.com/forecasting-market-movements-using-tensorflow-fb73e614cd06
• https://github.com/huseinzol05/Stock-Prediction-Models

Or you can try CPT Model: https://github.com/analyticsvidhya/CPT, so the algorithm can predict the next value based in the new data feed. Read more about it here:

• https://www.analyticsvidhya.com/blog/2018/04/guide-sequence-prediction-using-compact-prediction-tree-python/
• https://machinelearningmastery.com/sequence-prediction/

You should periodically (based on the range of normal fluctuations in the market, for example weekly) retrain the chosen model.