This isn't an exact science guys, it's a crapshoot that usually guarantees success based solely on probability. If you increase the size of the area of your guess, and the timeframe, you will always improve your odds. Earthquakes seem to follow a pattern, and rahul is probably basing his predictions on historical earthquake patterns. There is a gigantic amount of around 5-6 magnitude earthquakes all the time around the world. Check this list out:
Page Redirection . There is a large number of 5-6 magnitude earthquakes, especially off the coast of northern hokkaido or east honshu as well.
Anyways, predicting large, powerful earthquakes in the 8-9 magnitude is very difficult though. They seem to follow a pattern as well, but you can't get it down to a week timeframe, it is usually in the tens to dozens of years. The best method I see for "predicting" earthquakes is the new advance earthquake warning systems they have developed. The systems detect the "p" or "compression" waves after an earthquake hits. It is always the "p" waves that hit first, you cannot feel them, but they slightly precede the "s" or "shear" waves, the waves that you feel and cause the shaking. The further you are from the epicenter, the more advance warning you will get.
The quake warning system is going to be installed all across japan by this october:
MSN | Outlook, Office, Skype, Bing, Breaking News, and Latest Videos
http://mdn.mainichi-msn.co.jp/national/news/20070831p2a00m0na023000c.html
They are also planning to patch the earthquake warning data to keitais, which is an EXCELLENT idea in my opinion:
http://www.pcworld.com/article/id,132387/article.html