Can we use our historical data to predict the future? What if the historical data is random ? How can we see what is coming next 10 months ? Maybe AICE is a solution!
Motivation
I started my career in one of dreamiest countries in our planet: New Zealand. This is a country where you fall in love with nature and get addicted to lot of activities, now I am a surf and hike addicted guy. However if I want to enjoy the nature here, I have no way but rely on weather forecast, wave forecast, wind forecast and UV forecast. On the other hand, I created machine learning and data mining models for different industries such dairy, construction, energy and telco. They also starve for forecasting stuff, based on all their historical data they have collected over the years. I personally liked to get into stock market one day, but did want to spend lot of time analyzing multiple factors and technical indicators. Then I started to think: What if we have a model, which by some magic (not really magic, but comprehensive rules), uses only historical data, to predict something about future? I am not talking about stationary data, or data with clean and neat pattern, which are used by timeseries models . I am talking about random patterns. What if we have a model, which is capable in fishing out a pattern, out of a random pattern? I started to think of creating such a mode.
Convergent Evolution
In fact, once again, the nature gives us clue, how out of all random patterns in history of our planet earth, different creatures evolved over the time, to adopt them selves to survive. This process continued unlimited times over millions of years, till we see creatures which have inherited a lot from their ancestors, but they look so different and perfect nowadays, that barley we can relate them to their earlier versions. That is called convergent evolution. According to Wikipedia, Convergent evolution is the independent evolution of similar features in species of different periods or epochs in time. Convergent evolution creates analogous structures that have similar form or function but were not present in the last common ancestor of those groups. Does this concept resemble something in AI world ?
“Convergent evolution is the independent evolution of similar features in species of different periods or epochs in time...that have similar form or function but were not present in the last common ancestor of those groups.
AICE : Artificial Intelligent Convergent Evolution
In fact, ensemble models are very basic form of convergent evolution. What if like in nature, we create a multi-layer ensemble model, which can learn what is good and what is bad for its health, over the time, and improves it self? Well, I started to develop it. I call it AICE (Artificial Intelligent Convergent Evolution) model.
AICE can predict anything, using any given historical data. It automatically picks the good data, and predicts the target from 1 to 12 months ahead for monthly data. AICE uses sophisticated methods to prepare the data and derive hundreds of features out of just limited given features/columns. Then, it is doing exactly what we discussed so far. AICE is using very basic algorithms to begin with, and over the time it gets smarter and smarter, and adopts itself to most recent changes. It helps you to use your data up to now, and make prediction up to 12 months ahead, with high accuracy. You don't need to be worried about data preparation, model adjustment or data scaling. All you need is to have a historical dataset with a date column, and unlimited numerical columns. AICE is capable of making good predictions even using very limited numerical columns, because it is smart enough to derive multiple columns from every single given numerical column, and then chose the most influential factors.
In my next post, I will show you how AICE can help us to predict maximum/minimum stock price per month, from 1 to 12 months ahead. Stay tune.
AICE can predict anything, using any given historical data. It automatically picks the good data, and predicts the target from 1 to 12 months ahead for monthly data.
How to use AICE? Is it an enigma black box?
Yes and no. AICE is a sophisticated multilayer ensemble model with complicated adjusting factors. Not everyone knows how algorithms behind intelligent robots work for example. Sometimes companies don't want to release their secrets, and sometimes the process is too complicated and public wouldn't be interested. For now, AICE is a young algorithm, and I am working on it to make it better. Next update on AICE will be about empowering AICE using News or textual information updates. It is under development and early results look quite promising. If your company is interested in using AICE, please get in touch, I can start with free trial predictions. In my next posts, I will show how I used AICE to predict stock market price.
It sounds promising. Keen to see your new release on AICE. Good work