A systematic review on overfitting control in shallow and deep neural networks

Intelligent Transportation Systems (ITS) are much correlated with data science mechanisms. Among the different correlation branches, this paper focuses on neural network learning models. Some of considered models shallow and they get some user-defined features learn relationship, while deep extract necessary before by themselves. Both these paradigms utilized in recent intelligent transportation systems to support decision-making aid operations such as frequent patterns mining, regression, clustering, classification. When learners cannot generalize results just memorize training samples, fail necessities. In cases, testing error is bigger than error. This phenomenon addressed overfitting literature. Because, issue decreases reliability systems, ITS applications, we use over-fitted machine for tasks traffic prediction, signal controlling, safety emergency responses, mode detection, driving evaluation, etc. Besides, a great number hyper-parameters, more attention. To solve problem, regularized can be followed. The aim review approaches presented regularize categories studies. Then, give case study that uses version convolutional (CNN).