Enhancing Fantasy Football Data Prediction Model

To help the visitors to our Fantasy Sports Fans website (fiso.co.uk) that play the Fantasy Premier League (FPL) game (which is played by over 8 million people worldwide) we have a webpage at http://www.fiso.co.uk/crackthecode.php that attempts to predict which footballers' prices will change within the FPL game. The more popular the player is (based on 'transfers in' or 'transfers out' by FPL managers) the more likely the player's price within the game moves upwards or downwards. It is a game where you are limited by a budget when buying players into your fantasy soccer squad and so it is useful to know when a player's price is likely to increase or decrease. The data is scraped by us every 30 minutes from two FPL webpages that contain all the necessary data (updated every 30 minutes). Our data prediction model is not as accurate as it can be (not all 'transfers' count towards the player's popularity, and thus price change, and this is where our model needs improving) but we're not sure how to increase its accuracy. We know it's possible as another website has a more accurate prediction engine (or engines) than ours.