IMPROVED MACHINE LEARNING METHODS FOR PROTECTING CLOUD-BASIS DATA

Abstract

Now more than ever, data processing and storage on the cloud must adhere to stringent security protocols. In this research, we look at how effective machine learning may be in protecting data stored in the cloud. Various machine learning models were tested in this environment in three separate experiments. A Random Forest model was used in Experiment 1, which yielded 95% accuracy, 0.92 precision, 0.96 recall, and 0.94 F1 Score. This demonstrates the model's ability to classify security risks with a reasonable ratio of correct to incorrect predictions. In the second experiment, the accuracy was increased to 97% using a Deep Neural Network (DNN). F1 Scores of 0.96, 0.98, and 0.94 for recall, precision, and otherwise show that the DNN can distinguish between threats and regular operations. A potent instrument for cloud security, the model accurately identifies intricate patterns. In Experiment 3, we presented security analysis that uses reinforcement learning, more particularly Q-learning. The model was able to identify threats with an 88% detection rate, however there was a trade-off between true and false positives due to its 0.05 false positive rate. The 0.12 false negative rate suggests that the accuracy of threat detection has been improved. Modern machine learning can secure data stored in the cloud, according to these findings. Models trained using Random Forest and Deep Neural Network provide excellent accuracy while maintaining fair precision-recall trade-offs. In contrast, Qlearning-based reinforcement learning shows potential but requires tweaking to enhance the model's accuracy and false positive rates. Although there is an ongoing need to adapt and learn in response to emerging risks, the model should also address security requirements. An adaptable and secure cloud computing infrastructure is better able to withstand change, according to this research.