Soelaiman, R., Pratama, Y.H., Subakti, M.M.I. dan Purwananto, Y. (2020) Genetic Algorithm Approach for Automated Generation of Neural Networks Architecture for Robust Digit Recognition


Digit recognition is a special part of research in Optical Character Recognition. It is a common technique to recognize the numeric characters from printed images. A solution’s design for the digit recognition problem on a case study of SPOJ Hard Image Recognition (HIR) has been proposed in this paper. The case study’s problem has challenging constraints such as runtime limit and source code limit and it has many noisy images. An artificial neural network has been implemented to solve this problem in consideration of its simplicity yet powerful enough algorithm. The selection of best ANN architecture is commonly achieved through trial and error process, which is a very time-consuming process. This paper also provides the use of a Genetic Algorithm to determine the architecture of ANN automatically. The creation of the dataset also has an important role to improve the accuracy. The proposed architecture development successfully passed the challenging constraints and achieved a high score of 108 at SPOJ HIR. The score obtained by using GA is higher than our predetermined ANN architecture.


Artificial Neural Network, Digit Recognition, Genetic Algorithm, Optimization, Pattern Classification.