So what’s the best learning method, by scientific research result?

tl;dr: Of the learning strategies that deserve consideration, the two most effective choices were practice testing and distributed practice(or spaced repetition).

While many different approaches are utilized, there are really only few that are effective. The following is not compiled of truisms your grandmother might throw at you; instead it is based on the scientific results found by Dunlosky et al (2013), which reviewed nearly 400 studies from the 60s to the present.
First, let us take a look at elaborative interrogating. This method is essentially the academic application of a child continuously "Why?" When employing elaborative interrogating, learners aim to come up with explanations for facts, rather than simply memorizing them. The positive outcome of this lies supposedly in the more effective integration of new knowledge, provided through the process of asking for causality.
In the above-mentioned study, this technique scored a moderate success rate. While applicable to a wide range of topics, it remains questionable to what extent the effects of elaborative interrogating can be measured concerning lengthier subject matters. Moreover, the research partly suggests that the positive results reached by using this technique may be limited to students with a low knowledge domain.
A similar technique, that is likewise known to be employed by many students is self-explanation. This method requires one to explain one’s procedure when problem-solving. For example, students solving mathematical equations would be asked why they are doing what they are doing at every step. Like elaborative interrogating, supporters of self-explanation argue that the acquisition of new knowledge is enhanced when mixing it with previously known information.
After testing, self-explanation was graded moderately successful for many of the reasons illustrated above. While learning outcome proved to be highly positive, research concerning the longevity of the memorization of new information is yet inconclusive. An additional drawback of this technique is the substantial time demanded – especially when faced with the somewhat shaky results.
Especially when faced with lengthy texts and many ideas to comprehend, students will often rely on summarization in order to grasp the new information. This – as far as the submitted science is concerned – consists of reading the text and then writing a summary, listing all the important facts and thoughts comprised within the text.
The results for summarization might seem counterintuitive. Surely, almost anyone has been told to employ this method at one point or another. However, researchers ranked its utility as low. Although summarization can be useful for those already well-versed in applying this technique, most people are not very proficient at it and require extensive training. Moreover, it is yet unclear how much success is impacted by factors such as length of text. And finally, while there are studies examining summarization training in the classroom, no substantial studies have looked at the actual success-rate students can expect from utilizing this approach.
Perhaps the most common of all learning strategies is underlining or highlighting (both will be treated as equivalent). This entails students marking new texts when reading them in order to grasp new information immediately. The great appeal of this strategy lies in its simplicity and minimal time requirements.
Contrary to popular belief, marking texts in some fashion proved highly inefficient. Not only does it do almost nothing to boost performance, it can even hinder progress with complicated texts that require concentration and inference on the part of the learner. Since this technique is usually employed with previously unknown topics, students will not know what to highlight and end up stifling their own progress by, in essence, marking random and unimportant passages.
Going back in history, no technique has stood the test of time like mental imagery. Initially rising to prominence in ancient Greece, it is still popular to this day. Since mental imagery is an enormous field in its own right, we will focus on two variations: creating keyword mnemonics when learning a foreign language and mental imagery to help understand difficult texts.
Keyword mnemonic proved virtually useless, as it is only useful for mnemonic-friendly vocabulary. Students who utilized this method faired no better than a control group simply committing the vocabulary to memory. The story is slightly different for mental imagery. However, although there has been some promising work done on the matter, which indicates possible advantages, it seems that mental imagery is likewise bound to topics that can easily be converted into images. Is should be noted that mental imagery has not yet been sufficiently studied as a method of memorization and could very well turn out to yield positive results. For now, however, both techniques rank as low utility.
Another highly popular method for learning is rereading either chapters or entire textbooks. This approach is popular with students of all academic levels. When asked, 55% of students at an elite university (average SAT score above 1400) reported regularly rereading material in order to put it to memory.
Although rereading appears to offer some utility, the generality of this has not been well established. Furthermore, while rereading is relatively economic in terms of time spent, it is also highly ineffective when compared to other learning methods. For this reason, rereading was ranked as having low utility.
A far less popular (sounding) technique for learning is practice testing. Most students will associate the term with high-pressure situations in classrooms. However, this is not what researchers evaluated. Practice testing - in the sense of this study - is any kind of practice that students can engage with on their own. This entails flashcards, tests at the end of textbook-chapters or any other method of the sort.
Perhaps surprisingly, practice testing was ranked as having high utility. During the research conducted, it was found as having a wide range of applications; suited for learner of all ages and virtually any topic. It also ranked high in terms of recallability when students are asked to reproduce the acquired knowledge later on. As such, practice testing is one of the most promising techniques for learning.
Often, students will encounter new information multiple times, perhaps first in the classroom and later in their textbook. Therefore, scientists have looked into the effects of distributed practice (or Spaced Repetition) on learning. This means stretching the learning process over a longer timeframe, with the amount of time invested being equivalent to those studying once or twice in longer sessions.
The results were astoundingly in favor of this method. It proved to be effective for learners of all ages and applicable to any sort of learning material. While there are no conclusive studies demonstrating the effectiveness of this technique on very complicated topics, the already conducted research suggests results should not vary from those observed. Because of this, it is ranked as high utility.
Virtually all academic topics can be further divided into numerous subtopics. This in turn poses the question of how to study. While the traditional approach has been to blocking subtopics – that is, to study one topic with all its problems and questions one day and proceed to the next topic on day two – a contrary method has recently risen to prominence. Interleaved practice asks students to constantly switch between subtopics.
The paper found interleaved practice to have moderate utility. While it does seem to have a relatively dramatic effect on mathematical skills, no research has been done on other topics. Additionally, although the sample size of literature on this specific technique is small, it contains a worrying amount of null effects. Whether this can be contributed to its ineffectiveness as a whole, or if we simply do not understand interleaved practice well enough yet, remains to be seen.
The paper shows that many of the conventional learning methods (such as highlighting and summarizing) are widely ineffective, while others, that remain mostly overlooked (especially distributed practice and practice-testing) proved to have high utility. Although the results might seem counterintuitive, they are supported by hard facts and should therefore be taken seriously.