With the great industrial expansion and development in recent times, appropriate tools and systems need to be used to ensure the proper functioning and operation of these systems. One of most commonly used systems is the Supervisory Control And Data Acquisition system (SCADA), it provides supervision, control and monitoring of the system through the use of an extended network of sensors and actuators along with programmable logic controllers managed through Human Machine Interface devices. All of these components are connected using an I/O network that is usually connected to the internet for easier access, this makes the system susceptible to malicious network attacks. Adequate protective measures needed to be implemented to protect these systems from attacks as the successful execution of one can have devastating effects. In this work we propose an attack detection model based on the use of deep learning autoencoders to help detect complex malicious response injection attacks on SCADA based gas pipeline system using a pre-existing data set. The proposed method has shown a high detection rate value of up to 96%, which is considerably higher than the existing methods. The method relies on the use of the squared mean error value for the predictions made using the autoencoder model that was trained using only the normal data instances from the data set.