For those wanting a more advanced approach than RSI to waveform analysis: http://www.vallen.de/products/software/wavelet
I believe the conclusion that Wilder's RSI has close to zero capacity to predict future market behaviour can be applied to all technical indicators. But for historical analysis of price, they are just fine (hence the term Technical Analysis, not Technical Forecasting). My admittedly quick read of the paper you attached on wavelets leaves me with the impression that the price series is first decomposed to remove the trend component and the waveform/cycle analysis is applied to the transformed series. Of course, RSI and other traditional indicators are usually applied to the raw price series which contains both trend and cycles mixed together. Did I get this right, or am I missing another key differentiator?
Yes, that is correct....however if you are using SVM, Deep Learning, nets etc. tossing in both can't hurt as the analytics process, at least with SVM, will tell you which works best. The use of raw TI is in essence a religious ritual similar to lighting a votive candle to Saint Joseph....not much science to it but if it feels good then enjoy!
You might consider a bit of study in physics or related fields but the short answer is to decompose the waveform (market) into it component forces: sort, long term, trend, reactive impulsive, random noise, then forecast each of those (the ones subject to predictive modeling) and then merge the forecasts to actually trade the market.
TA in general will only have 50% precision at best if price moves are random. I still use TA in order to give some kind of context to the meaning of high and low, and in order to get at least slightly higher odds than 50/50 to a price move. As for RSI specifically, I would urge folks to read John Ehlers work on applying a roofing filter first to price before calculating he RSI. A roofing filter is just a two pole highpass filter and two pole supersmoother filter combined.
The qualification is predictive capacity in the sense of an objective predictive model in the terms of formal analytics. I have no doubt that those still making trading decisions manually have found helpful TIs.