You can consider biological traits like gender, age, physical development, physical disabilities, health, motor skills, coordination, and diagnosed learning disabilities. Moreover, thinking skills, reading skills, attentional focus, past success, oral language development, written language, proficiency with sequencing, proficiency with categorization, and proficiency in identifying logical arguments are also important. interest and levels of student motivation, achievement, productivity, and perseverance are recommended.

thank you dear Ajit kumar Roy , that certainly gives me a lot of clues. i wonder if there has been any works on a categorization of these features, in a sense that which ones are observable and detectable sooner ( in a formative way).

Because if i choose such features (traits) the ML algorithm can be used as a mixed approach to clusters and do formative assessment on students in traditional class rooms, and therefor in less developed areas which is ultimately my goal.

Iman, if you have a traditional classroom, the features that you may consider are. Previous scores (You may even consider schooling scores) Gender (M,F) , Urban, rural. Using too many features may confuse you during analysis and not considered good practice.

this article might help you

https://pdfs.semanticscholar.org/0f43/45d8d809f3952c4b6104f7cba462b398b7d3.pdf