The rapid rise of malware threatens global cybersecurity, demanding robust detection of known and unknown threats. Traditional machine learning struggles with imbalanced datasets, yielding biased models with poor malware detection rates. MalwarNet, a pioneering framework, addresses this through a novel SMOTE-KNN approach, integrating Synthetic Minority Over-sampling Technique (SMOTE, k = 5) with KNN boundary filtering. Its pipeline includes preprocessing (feature extraction, imputation, normalization), balancing, scaling, splitting, training six algorithms (Gradient Boosting, Random Forest, Decision Tree, KNN, AdaBoost, Logistic Regression), and evaluation via accuracy, precision, recall, F1-score, and ROC-AUC. On two public malware datasets, it achieves 98% accuracy and 99% recall, excelling in finance, healthcare, and critical sectors.