Neural network approach to Russian botnet
This paper contributes to the ever-expanding body of literature on machine learning and predictive analysis. The research models the out-put of Twitter accounts associated with the Russian Internet Research Agency during the Ukranian Crisis. The paper sets a competitive fore-cast benchmark of 82.75% accuracy with vector autoregressive moving average model with exogenous variables and proceeds to employ recur-rent neural networks. The results suggest that the time series can be accurately predicted using various architectures, with bidirectional long short-term memory variation achieving a 92.45% forecast accuracy. The predictions are in part based on tweets’ content features extracted with natural language processing.