In a fiber loop ring-down spectroscopy (FLRDS) system, the gain saturation absorption effect of the erbium-doped fiber amplifier (EDFA) leads to gain fluctuations in attenuation pulses thereby significantly impacting measurement accuracy. To reduce this impact, we propose an end-to-end model using one-dimensional convolutional neural network (1D-CNN). By preprocessing FLRDS data and feeding it into the model for training, the model can predict data with unknown concentrations. This method avoids the traditional decay curve fitting process and indirectly reduces the negative impact of EDFA gain saturation absorption behavior on FLRDS measurement accuracy.