I assume you built a neural network (or similar) system that tries to predict tomorrow's moving ave slope. Its easy to get a relatively high % forecast accuracy. In a simple moving average, all 10 inputs carry the same weight. Your model already knows 9 of the inputs and it also knows the 1 input that is dropping out. Its easy for the model to compare the 1 dropping out to the 9 that are staying and make some guesses about whether the slope will be up or down. While the model will likely have a very high accuracy, it does not actually mean much in real world use. If your model could tell me if tomorrow would be up or down, then we have something, but the slope is not that helpful in this case unless it was close to 100%. Then you could use that information to determine a "range" the stock would have to close at to produce the predicted slope. Then from there you could develop several different classes of strategies using that information.... But I am pretty certain that once you remove the "predictive edge" you have just due to knowing 9 data points,the system is not that much more accurate.

frostengine, have try the range you talking about but the range is quite big and not that helpful to trading suppose the slope is up the close may also close quite low compare to previous close and slope is up any other use of the big range?

One could calculate the angle of the linear regression line from the last pivot point. And therefore the extension of at least 1 bar. Useful until the next pivot point ?