LSTM Model to Forecast Time Series for EC2 Cloud Price

With the widespread use of spot instances in Amazon EC2, which utilize unused capacity with unfixed price, predicting price is important for users. In this paper, we try to forecast spot instance price by using long short-term memory (LSTM) algorithm to predict time series of the prices. We apply cross validation technique to our data set, and extract some features; this help our model to learn deeply. We make the prediction for 10 regions, and measure the performance using root-mean-square error (RMSE). We apply our data to other statistical models to compare their performance; we found that our model works more efficiently, also the error is decreasing more with cross validation and the result is much faster. Based on our result we try to find the availability zone that less become over on-demand price and less changing price over time, which help users to choose the most stable availability zone.