LSTM-ANN: A Hybrid Deep Learning Model for Task Failure Prediction in Cloud Computing Environment

Abstract

A cloud computing environment can perform millions of tasks per day, and reliability and efficient use of resources are significant issues for cloud service providers. Task failure due to workload imbalance, resource contention, hardware failure, and abnormal task execution behavior can have a dramatic impact on system performance and negatively impact on SLAs. In this research, we have proposed hybrid deep learning model for task failure prediction namely LSTM-ANN. The simulation was performed in Python language and Google Cluster Traces 2019 dataset was used for performance checking. The proposed framework compared with Artificial Neural Network (ANN), Long Short-Term Memory (LSTM), and Hybrid GRU-ANN, to perform binary classification of failed and non-failed cloud tasks. The study involves data preprocessing, feature selection using SelectKBest, ANOVA F-test scoring function, and workload-based event transformation in order to enhance the performance of the prediction. Evaluation of the proposed model was performed using accuracy, F1-score, ROC-AUC, RMSE, and confusion matrix analysis. The experimental results showed that the Hybrid LSTM-ANN model had the highest overall accuracy of 96.08%, F1 score of 0.9173, ROC-AUC value of 0.9925, and the lowest RMSE value of 0.1721. The results of the Hybrid GRU-ANN model were also very competitive in terms of performance but not accurate as the LSTM-ANN model produced.