Neural Network – A Black Box Model

• Main Authors: Kuok, Kuok King, Chan, Chiu Po, Md. Rezaur, Rahman, Khairul Anwar, Mohamad Said, Chin Mei, Yun
• Format: Book Chapter
• Language: English
• Published: Cambridge Scholars Publishing 2024
• Subjects: TA Engineering (General). Civil engineering (General)
• Online Access: http://ir.unimas.my/id/eprint/45998/1/978-1-0364-0804-6-sample.pdf; http://ir.unimas.my/id/eprint/45998/; https://www.cambridgescholars.com/product/978-1-0364-0804-6

Artificial Neural Network (ANN) is a computational model based on the structure and operation of biological neural networks. It is a black box model due to its complexities and difficulties in understanding how to make decisions and predictions with complicated internal structures and huge parameters involved. The basic unit of ANN is the artificial neurons. A group of neurons forms a layer. There are three layers in ANN, namely, the input, hidden, and output layers. Forward and backward propagation are two common learning processes adopted for adjusting weights and biases in ANN. Various activation functions are used, such as Hard limit, Tan-Sigmoid, Linear, Log-Sigmoid, Rectified Linear Unit (ReLU), Hyperbolic Tangent (tanh), and Softmax, enabling ANN to simulate complicated relationships and perform nonlinear transformations. Three learning paradigms of ANN include supervised, unsupervised, and reinforcement learning. A variety of metaheuristic algorithms have been used to train ANN, including Genetic Algorithms (GA), Particle Swarm Optimization (PSO), Simulated Annealing (SA), Ant Colony Optimization (ACO), Tabu Search (TS), and Harmony Search (HS). To date, ANN has been successfully adopted in streamflow prediction, rainfall-runoff modeling, groundwater modeling, water quality modeling, and water demand forecasting.