What does "state of the order book" mean?
Besides the best bid and ask, what other information does your data contain? E.g. depth of the book? If so, how many levels?
How it "return" defined? Buy at the best ask and sell at the best bid two minutes later? At this granularity, price impact may matter. Depending on order size and size of the queue, you may not be able to fill your sell order at the best bid.
Besides these context-specific questions, for forecasting price movement at this frequency you should start with simple models (e.g. ARDL models with appropriate LOB variables as regressors).
At this frequency, price movement is mostly due to trading algorithms reacting mechanically to the state of the LOB in some prescribed manner.
You may want to consider the two sides of the LOB separately. In that case a simpe VAR (multivariate version of ARDL) would be a good starting point.
(Consideration of filtering models would suggest you have reason to believe there are unobserved systemic driver among the trading algorithms.)
Relevant info on forecasting using AR/ARDL/VAR can be found in probably any introductory text on time series, as well as online.