The objective is to develop a higher-performing automated sleep staging system by exploiting multi-modal polysomnography signal recordings . Three modalities of PSG signals, namely electroencephalogram (EEG) and electromyogram (EOG) were considered to obtain the optimal fusions of the PSG signals, where 63 features were extracted. These include frequency-based, time-based, statistical-based, entropy-based, and non-linear-based features. We adopted the ReliefF feature selection algorithms to find the suitable parts for each signal and superposition of PSG signals. Twelve top features were selected while correlated with the extracted feature sets’ sleep stages. The selected features were fed into the one-dimensional convolutional neural network to validate the chosen segments and classify the sleep stages. This study's experiments were investigated by obtaining epoch-wise testing schemes. The proposed research was performed on the sleep-EDF(S-EDF) public dataset. In this work, we demonstrated that the proposed fusion strategy overestimates the common individual usage of PSG signals. The proposed experimental results reported an overall accuracy of 98.97%, 97.89%, 89.81%, 83.83%, and 83.72% for distinguishing between ‘rapid eye movement stage (REM) vs. non-rapid eye movement stage (NREM),’ ‘deep sleep (NREM-N3 + NREM-N4) vs. light sleep (NREM-N1 + NREM-N2)’, ‘wake vs. sleep (NREM + REM),’ and ‘wake, deep sleep (NREM-N3 + NREM-N4)’, ‘light sleep (NREM-N1 + NREM-N2), and REM’, ‘wake, N1, N2, N3, and REM’ respectively.