That paper is riddled with factual errors and typos. Frankly, I don't even understand what's going on in Figure 2. Conference looks shady.
Also, in other paper by the same author you have mentioned ("Identifying Tor Anonymous Traffic Based on Gravitational Clustering Analysis"): according to 2nd paragraph of Introduction, the paper is written with purpose of "catching criminals", rather than improving fingerprinting resistance of Tor. This might give another insight into their intentions.
I agree with Vini they probably haven't found anything. Furthermore, we seen much better accuracy with prior attacks.
As for blocking Shadowsocks(and other randomizing protocols, since being random is fingerprint in itself), you should be able to identify them with simple entropy+length tests(extremely strong feature authors didn't use for some reason) and get false negative rate close to 0, and false positive rate below 1%[1]. Apparently, this false positive rate is not small enough (see: base rate fallacy), and censors never employed it. Nevertheless, I believe those attacks might be improved with more heuristics and features to approach practicality, albeit not at line speed rate.
[1]
http://pages.cs.wisc.edu/~liangw/pub/ccsfp653-wangA.pdf - Seeing through Network-Protocol Obfuscation
Figure 2. Accuracy goes up with test set size for some reason, and accuracy rate in Chinese is linear, as opposed to English, I guess. Wat.