unfortunately, this is wrong. your code is subject to lookahead bias because you scale the entire dataset before training. furthermore, you possible have information leakage between adjacent train-test-folds. i am 100% sure that all the performance stems only from these factors. you can be lucky if you get sth like 50-53% afterwards. also, you need to compare your strategy to benchmarks like random strategy, buy and hold, as well as for insance the momentum strategy (i.e. invest according to the sign of the previous realized return). if i was you, i would add a remark at the start of the article for people who do not read the comments. even better; rerun the procedure correctly and report the actual results.

its not that easy!