This　paper　proposes　a　student　behavior　prediction　algorithm　based　on　MLP　and　adaptive　learning　rate　optimization.　The　algorithm　combines　the　powerful　nonlinear　modeling　ability　of　multilayer　perceptron　and　the　advantages　of　adaptive　learning　rate　optimization　algorithm,　and　can　effectively　process　the　complex　features　in　student　behavior　data　and　improve　the　prediction　accuracy.　A　student　behavior　dataset　from　an　online　learning　platform　was　used,　including　data　in　multiple　dimensions　such　as　learning　time,　number　of　interactions,　and　grades.　Experiments　show　that　this　algorithm　is　effective　in　predicting　students＇　behavior.　The　predicted　precision　of　the　algorithm　is　95.6%,　the　recall　rate　is　92.3%,　and　the　FR　is　94.0%.　Compared　with　traditional　machine　learning　algorithms　(SVM,　DT,　etc.),　the　precision　is　increased　by　6　percent,　and　the　training　speed　is　faster,　about　25%.　In　addition,　the　algorithm　showed　strong　robustness　and　generalization　ability　when　processing　high-dimensional　features　and　sparse　data.　This　study　provides　new　ideas　for　the　analysis　of　student　behavior　data,　especially　in　student　grade　prediction,　behavior　analysis　and　personalized　learning　recommendation.　It　has　important　application　value.　©　2025　IEEE.