Abstract

Non-productive task failures and unexpected delays in cloud environments may cause inefficient use of resources, and therefore less reliability for services. Catastrophic failures can be avoided by accurately predicting the outcome of each task performed by cloud resources. In this paper a comparative study is proposed on the application of deep learning models to predict the status of a cloud task based on the Alibaba Cluster Dataset. The subsets used from the dataset were used for preprocessing and feature selection, resulting in a balanced dataset with 112,928 task instances and four task status classes for training and evaluation of the model. The study compares the performance of recurrent models, hybrid models, and transformer-based models such as LSTM, ANN, GRU, GRU-LSTM, GRU-ANN, Temporal Fusion Transformer (TFT), Informer, and Tab Transformer. The results of the experiments revealed that the GRU model achieved the best precision of 97.3%, while the LSTM models and the ANN models showed similar performance. Transformer architectures performed competitively but did not achieve the highest accuracy among the various models tested, with Tab Transformer being the lowest of the lot.