An automatic sleep monitoring system is a prime requirement in order to minimize analysts' workload of visually inspecting large-scale data for sleep scoring and to improve the accuracy at large range. This research introduces a robust, precise single channel EEG-based automated classification scheme for sleep states based on Variational Mode Decomposition (VMD). After pre-processing and decomposition, various features are obtained from the decomposed modes and the pre-processed EEG signal. This proposed method aims to use VMD for the first time in EEG based sleep scoring to the best of our knowledge. Our model achieved its best performance using 10-fold cross-validation in RF classifier after enormous experimentation. Our model yielded promising classification performances of 98.1894%, 95.4183%, 93.2352%, 92.0181% and 90.8283% overall accuracy for 2-states to 6-states classification respectively. This scheme also shows a high or comparable detection accuracy for awake, S1, S2, S3 and REM stages of 5-stage sleep scoring.