Spotting indications of unsafe human behaviour, a leading cause of an accident, is critical in providing a safe workspace. Among factors affecting human behaviour, stress and overload are the most significant ones, where limited knowledge exists on their underlying causes. Tracing physiological signs caused by stress and overload might be a feasible approach in detecting a specific neurological status leading to unsafe human behaviours, such as disobeying safety rules, standards, and instructions. In this paper, we present a deep learning technique to recognize distinctive neurological status by assessing physiological signals such as temperature and heart rate. An open database of non-EEG physiological signals was used to train and test the model. The database includes electrodermal activity, temperature, acceleration, heart rate, and arterial oxygen level signals of 20 healthy subjects through relaxation, physical, emotional and cognitive activities. A robust automated pattern recognition method, using deep learning, was used to predict and identify stress and overload. The experimental results indicate that the model can detect neurological status with higher accuracy than the traditional classification-based methods.