Developing　proactive　prevention　measures　before　accidents　is　a　powerful　means　of　kerbing　traffic　accidents.　Many　studies　have　confirmed　that　the　state　of　the　traffic　would　fluctuate　significantly　before　an　accident,　representing　as　traffic　instability.　This　study　develops　a　Traffic　Instability　Index　(TII)　to　evaluate　and　identify　factors　affecting　traffic　instability,　utilizing　Non-negative　Matrix　Factorization　(NMF)　based　on　aggressive　driving　behaviors　and　traffic　flow　data.　Machine　learning　algorithms　and　Partial　Dependency　Plots　(PDP)　reveal　that　traffic　status　significantly　influence　TII,　followed　by　road　attributes.　The　findings　show　that　high　TII　is　linked　to　more　fatal　accidents　and　that　traffic　instability　is　adversely　impacted　by　increased　speed　differences,　traffic　volume,　and　the　number　of　entrances　and　exits,　but　reduced　by　the　proportion　of　trucks　and　median　openings.　Vehicle　type　mix　has　a　non-linear　effect　on　TII.　Additionally,　median　openings　mitigate,　while　entrances　and　exits　exacerbate　the　adverse　effects　of　vehicle　mix　and　truck　proportion　on　instability.　©　2025　Tongji　University　and　Tongji　University　Press