Machine learning to select features to create predictive model

Sorry if this is a little confused!

I am experimenting Forex data using keras to attempt to find phenomena that predicts a rise or fall in a forex symbol’s price. My dataset is millions of prices per minute per symbol with 50 or so technical indicators for each minute. So partly this is feature selection but I expect that within each feature there is a limit range in which the price is predictable.

I realise that in the vast majority of cases the forex price cannot be predicted from the data so simply feeding all the data into a neural network won’t work. However, I want to find out if during particular phenomenon it can be predicted. For example, if x feature is above 70 for 3 minutes, and y feature is increasing (and feature a is.. and feature b is… etc) then this is predictive of a rise in forex price. At the moment, I am attempting a brute force approach by randomly selecting the criteria for the features, then finding the data that meets that criteria then feeding the data into a neural network. But it isn’t proving very successful.

Can anyone point me in the right direction please? How would I begin to create a model that selects bands of the features that lead to a successful prediction?