I guess Jayachandran is using the TLC272. This chip offers an input bias current of less 600pA at 70°C, which doesn't seem to responsible for his observed DC offset shift. The offset voltage drift of typically 1.8µV/°C cannot cause the drift either. Nevertheless I would recommend to use the TLC277 instead of TLC272, because it offers a much lower initial offset voltage (500µV instead of 10mV).

Although input bias current of TLC272 is ultra low, a choosen smoothing resistor of 3.3MOhm is much too high. This results in a very high impedant node directly at input of OPamp, which becomes extremely susceptible against all sorts of interference. Choose a smoothing resistor in the 10...100kOhm range.

I agree with Rob, who says that 17mV offset voltage drift is caused by the 5V voltage regulator itself. It's not easy to obtain much lower offset voltage drifts, a few millivolts cannot be avoided. But you can relevantly improve your application by the following two simple and cheap options:

1. You could use the LM723. This regulator is ultra stable. (As ultra as they could make it, when they designed this chip some decades ago...)
Have a look at figure 5 of

http://cache.national.com/ds/LM/LM723.pdf

Take care, the LM723 needs at least 9V (better 10V) at input to provide its superior performance.

2. You can use a combination of LM317 plus LM336-2.5 as shown in figure 00571513 on page 7 of

http://cache.national.com/ds/LM/LM136-2.5.pdf

Kai