AboutGeorge Moustris Expertise I can answer most of the questions about the workings of a Fuzzy Logic machine, from the basics up to a certain advanced point.Questions regarding applications of Fuzzy systems intregrating with other technologies such as Neural Nets or Genetic Algorithms are propably out of my reach although,with some research,i could try to answer them(i have working knowledge of Neural Nets an Genetic Algorithms but i`m far from being consider an expert to these areas).
Experience I have experience in Fuzzy Logic gained in projects i have been part of.The last project was a robotic vehicle that used a Fuzzy Logic path tracker which i designed.
Organizations Aristotle University of Thessaloniki,M.Eng. Electrical and Computer Engineering , National Technical University of Athens,Phd, Computational Intelligence in Robotics (currently pursuing)
Question We are working on a project that we hope a computer programme will assist us with. It involves the outcome (predicted probability of an event) based on a sizeable amount of existing data from previous events, and might be used, for instance, to predict the probability of a vehicle accident on a given road, or a horse winning a particular race.
For our project we are using horse racing. So, if horse A wins race B and we have the data previous races that A ran, the parameters that are likely to see a successful race might include length of race, location, weather, jockey, etc.
Importantly, all of these parameters interact to determine the success of A. We have 5 years of data that includes the parameters that affect these races as well as the outcome of each race.
Our question is: Do you think that it is possible for a computer programme to be developed that will learn from this race data (possibly using AI and Neural Networks)? If we use existing data from Years -5, -4 and -3, and then use the computers predictions for the outcome of races at Years -2 and -1 these could then be validated as we already have these results. Could it then (itself) “fine-tune” its own predictions and ultimately “learn” to forecast the possible results over the following year (Y +1)? If this is possible how do you think we might achieve this?
Answer Hello Stephen,
to cut a (very) long story short, yes, such a programme can be developed. Typically such programmes implement some kind of Artificial Neural Network along with some data mining and manipulation algorithms (e.g. support vector machines, PCA etc). Forecasting is a field on its own and there is a vast amount of knowledge pertaining to it.
A place to start could be Cohonen networks, LINE and ADALINE algorithms and recursive networks. I suggest you get acquainted with them and see how to fit the to your project. Of course a general knowledge on forecasting would greatly aid your task.
