How Cindicator increased forecast’s accuracy by implementing its first Neural Network
We can now announce that Cindicator has deployed its first fully functioning neural network. Our development team has integrated the network into the Hybrid Intelligence ecosystem, and the first indicators based on the neural network are set to launch in the near future. “What does this mean?” you might ask, “And what are neural networks anyway?”
We’ll get to that.
When making predictions, one method of crowdsourcing opinion is to deploy the wisdom of the crowd. This means that, looking at a diverse range of opinions on a given issue, you take the ‘average’ response, in theory giving a more accurate prediction than any individual — no matter how knowledgeable — could provide.
At Cindicator, we went further and used a diversity of Data Science methods, incorporating statistical science and machine-learning methods to improve the accuracy of our predictions. For example, we used an approach based on the Hidden Markov model, modelling probabilistic Markov chains on the basis of previous errors made by users. Taking into account the history of their predictions (and errors) we were able to assess the likelihood of future predictions being correct.
Another approach is Bayes’ theorem, which allows data scientists to look beyond precedent and to actually analyse the environment in which the events in question are taking place. Using the theorem the likelihood of an event based on a multitude of conditions related to the event is calculated. In our case, we looked at different statistical metrics like standard deviation, skewness, heteroscedasticity, etc. for each user, allowing us to view all users not like a single answer, but as a distribution pattern and visualise overall and individual tendencies to error.
Our results can also be streamlined by looking at a given stratum of high-performing users rather than our entire user base, and there are various ways of deciding the cut-off point for the sample group.
All this given methods incorporated into our ML pipeline in the form of a models. In order to improve the overall accuracy of our signals we have begun tests on a method for prediction incorporating a neural network. The neural network as universal approximator, locates complex, nonlinear dependencies between the models and gives an individual weighting to each model in order to arrive at a final prediction. It also, applying them in practice and further improving the accuracy of its predictions. Symbiosis of the AI and collective intelligent opens huge area for developing and continually improvements. This means that we will soon be able to offer even better indicators, increasing both the utility of our indicators and prospects for future product development.