Over the past decade, multivariate time series classification has been receiving a lot of attention. We propose augmenting the existing univariate time series classification models, LSTM-FCN and ALSTM-FCN with a squeeze and excitation block to further improve performance. Our proposed models outperform most of the state of the art models while requiring minimum preprocessing. The proposed models work efficiently on various complex multivariate time series classification tasks such as activity recognition or action recognition. Furthermore, the proposed models are highly efficient at test time and small enough to deploy on memory constrained systems.