These days I was focused on a personal project to provide public bitcoin forcasts using data mining algorithms (neural nets for the moment), updated each 3 hours, with a visual clue about the performance of the algorithm. My goal is to know how state-of-the-art algorithms work on such high-volatile trades.

For the moment I am only working with Mt-Gox and USD, and the accuracy is not that good (around 0.6 in best periods), but I am looking to improve it. Any help or advices are welcome.

http://aquirin.ovh.org/bitcoins/Forecast2.html