This thesis aims at studying student performance prediction for different kinds of course implementations and using Machine Learning techniques applied to behavioral data collected from a Moodle-based check-in/checkout system to support self-regulated learning. Modules 1 and 2 targeted students who first scored under 80, and Modules 3 and 4 targeted those who initially scored under 70. This multi-module framework allowed rigorous examination of how predictive factors change across different performance cutoffs and instructional models operating in Moodle Learning Management System context. The technology infrastructure for this research is the Moodle-based check-in/checkout system, which becomes a structured two-phase assessment system in this work for recording diagnostic and summative outcomes together with the automatic logging of detailed behavioral data during learning.