Another aspect to be considered is how we go about developing our systems. We tend to use the last 40, 50, or 100 drawings to mathematically find a way to predict the next drawing. Only if, in the future, those same numbers fall into the same positions will our system perform as expected.
As the drawings continue, our numbers fall into other positions, other columns but our system remains static. To us it appears that our algorithym has gone "cold" whereas in fact our algorithym was never "hot".
IMHO, to create a usefull algorithym a developer must look past the numbers and their relative positions. Look instead to find what the numbers are telling us. Is the sum odd or even? Is there a repeat number? Was the first number odd or even? Was the last number odd or even? How many odd numbers were drawn?
Of course I have been wrong before ( just ask my wife ) and ( if I may be permitted a bit of scarcasism ) will probably be wrong again sometime in the far distant future.