This is an interesting one:
Facial recognition technology can expose political orientation from naturalistic facial images
Kinda damaging to our ideas of free will. 
As I see it, there is a fundamental flaw in the OP’s premise: the vast majority of people who commit crime are not people who typically commit crimes.
Most criminals are people who made a bad decision. They didn’t plan to become criminals, and would not usually break the law. Instead, they did something stupid, like getting into a fight or a car accident after a night of heavy drinking.
If your plan is to predict that kind of behavior by looking at somebody, you might as well ask if you can tell if somebody is prone to anger or rash decision-making based on their appearance.
Even with criminals whose crimes took some planning, the crime is usually one of opportunity. Say a person steals things from a store. They may have a pretty strict moral code in other walks of life. Just because they committed a crime doesn’t mean they consider themselves to be a criminal, and most of the time they aren’t.
Of course, there are those who are constant lawbreakers. Criminologists will tell you that these people derive from a multitude of factors, usually based on socioeconomic status. Since that status is not an immutable one, their own particular appearance shouldn’t be a marker for their behavior, either.
(Although, in terms of how one’s behavior can impact one’s appearance, there’s the example of drug use. Prolonged use can certainly age a person).
So is Acupuncture; so is Astrology. Popularity is not a measure of whether something is supported by evidence. Do you want to talk about evidence-based science, or do you not? You say you do, but you behave as though you don’t.
‘All the proper evidence’ is doing some seriously heavy lifting here. There are scattered examples where appearance means something, sometimes, maybe, but this is far from sufficient evidence to support the graduation of physiognomy from pseudoscience to actual science, and there have been centuries of opportunity *in ‘western’, evidence-based science for this to happen. It hasn’t.
Then maybe don’t say things like:
Utterly irrelevant attempt at ad hominem, but that said, I look forward to watching the development of any Good Things About Hitler thread you might fancy starting.
I do too. Including eugenics.
I would be very interested to see a study where people were asked to pick criminals from a set of photos composed of:
- Mugshots of innocent people falsely arrested and/or falsely imprisoned
- College yearbook photos of people who went on to commit violence or murder.
You’re welcome to link some research papers here that say in their conclusion that there’s no correlation, or better yet, negative. Note that only negative ones can disprove a theory; blank ones just don’t reinforce it, or at best, make it more unlikely.
You’re the star again! If we accept a very reasonable and likely position that none of the newborn infants have a political tendency, then the result more or less proved that changes in our lives reflect on faces.
You’re arguing about the intention of crimes (or, typically, non-crimes) being more important than the act. As I see it, world’s justice systems predominantly judge based on actions. Moreover, most murders are crimes of passion, i.e. they fall into your category of “people who would otherwise not commit it and live an exemplary life”, yet they’re punished severely by society.
But I digress. What I did in the OP is to take a specific example and try to morph it into a broader question. I’m not interested in taking a criminal’s circumstance into consideration, that’d complicate things too much. If they’re widely identified as criminal, then we should use the label and focus on how to recognize their face. But then again, criminality is not the main topic, it’s the face.
But what if a person’s circumstance determines whether they are a criminal? While it might complicate things, if you are looking for a predictive measure it may be the feature you need to study. Certainly, most modern criminologists would tell you that it is the most insightful thing to consider.
That gets back to my original point: who is “widely identified as criminal”? As I said earlier, most people who commit crimes aren’t fairly identified as criminals, for the vast majority of their lives.
Are you then referring to a particular subset of people who people who commit crimes? If you are, then there is copious research that points to socioeconomic factors to explain that behavior. So, again, your intention to connect their faces with their behavior is spurious; there aren’t facial features that typically connect to those variables.
Now, lifestyle does impact one’s appearance. So, if you were trying to draw conclusions about how one lives from what one looks like, I bet you could make some realistic guesses (I.e. people from poverty would tend to look more beaten down than people from privilege). And, from there, you could perhaps make guesses about their behavior. But these would be tenuous guesses, and I doubt they would hold up under intense scrutiny.