A Novel Approach on Error Detection and Correction using Feedforward Artificial Neural Networks

Abstract

According to a Forbes survey around two and a half quintillion bytes of data is being created on the internet each day. With such a rate the amount of data to be structured is going to double in a time span of nearly one and a half years. It is not only the amount of data but the number of systems that communicate the data with each other increasing the complexity even further. With such a huge combination of data exchange possible, there is a high chance of erroneous data communication between communication channels. Though there are some hardware technologies like optical fiber which reduces such errors the cost for such systems is very high and impractical to implement on a full hundred percent scale. Moreover, with the current technology of error correction and reduction, the system is not immune to a large number of changes in the communication channel. If the number of bytes changes either in the data word or the appended code word the whole algorithm remains vulnerable. Hence in this paper, artificial neural networks are used to reduce the errors and make the system immune to a high number of changes in the data word or even the code word. This is achieved using artificial neural networks as an encryptor and doing overfitting of the data on the bytes. Since composite functions are created for every hidden layer of the neural system, therefore the change if a higher number of bytes does not affect the neural network output considerably and this forms the crux of this paper.