According to recent reports, at least 30% ofstudents at various colleges and universities areintermittently failing their courses, making academicunderachievement and student dropout and withdrawalsignificant problems for higher education and learningglobally. It is practically impossible to identify at-riskstudents early enough to intervene in time for it to matterbecause the traditional estimating predictors andpossibilities have very retrospective considerations. Theseestimates are all derived from exams, teacher judgmentthrough the use of manual assessments, and end of termresults. To identify at-risk students early and providepersonalized education and learning interventions, the studyproposed a data analytics-based proactive approach, whichincludes data-based explanatory artificial intelligence (XAI),machine learning (ML) and education based data mining(EDM). Along with datasets such as the Open UniversityLearning Analytics Dataset (OULAD) for benchmarkingpurposes, the framework also utilizes data collected fromother sources including academic performance, behaviorrecords, and engagement logs. Random Forest, XGBoost andDeep Neural Networks were utilized for the predictivemodeling, and the prediction power of the models can beexplained through SHAP-based explainability. Finally, theframework completes the cycle not only of predictions andtargeted intervention in the forms of adaptive tests,personalized study materials and early warning to parentsand teachers. Thus, preliminary testing indicates that inaddition to using intervention, the framework can boostengagement and pass rates by 15%-20%, will identify at-riskstudents as early as the first four weeks of the semester, andachieve over 85% accuracy on prediction.