Identifying malware, especially zero-day exploits, continues to pose a significant issue in cybersecurity. Conventional signature-based approaches are inadequate for identifying new threats, as they rely on established patterns of behaviour. To enhance the precision and effectiveness of anomaly-based malware detection, this study examines the application of hybrid machine learning (ML) approaches trained on the AWSCTD dataset. By utilising sophisticated feature selection algorithms and incorporating metadata, this study demonstrates notable improvements in detection rates while reducing false positives. Comparing with deep learning models reveals the trade-offs between computational efficiency and accuracy. The BestFirst-SVM method, a hybrid machine learning technique that combines the feature selection capabilities of BestFirst with the classification power of SVM, outperformed other traditional machine learning techniques with an accuracy of 97.35%. A thorough summary of recent developments in the field is also provided, including insights from research articles published in respectable publications.