Optimization of Memory Management Using Machine Learning
Presentation Type
Oral Presentation
Mentor/Supervising Professor Name
Alférez, Harvey
Presentation Location
On Campus
Description
Memory overload can cause undesirable behaviors in a system. Proactive actions for memory safety may address this problem. Our contribution is to use machine learning models to classify different states of system memory using a dataset collected from a Raspberry Pi device. Several experiments were done on three datasets. For the first dataset, k-nearest neighbors had the best F1 score to classify medium and high RAM usage classes. For the second dataset, an artificial neural network had the best F1 score for each class. For the third dataset, logistic regression had the best F1 score for each class. The sliding windows used for classifications were created using inputs of 10 seconds of memory data usage to predict the next second of usage. This approach could eventually be used to classify and prevent memory overload scenarios.
Optimization of Memory Management Using Machine Learning
On Campus
Memory overload can cause undesirable behaviors in a system. Proactive actions for memory safety may address this problem. Our contribution is to use machine learning models to classify different states of system memory using a dataset collected from a Raspberry Pi device. Several experiments were done on three datasets. For the first dataset, k-nearest neighbors had the best F1 score to classify medium and high RAM usage classes. For the second dataset, an artificial neural network had the best F1 score for each class. For the third dataset, logistic regression had the best F1 score for each class. The sliding windows used for classifications were created using inputs of 10 seconds of memory data usage to predict the next second of usage. This approach could eventually be used to classify and prevent memory overload scenarios.