A Risk Estimation Framework

Step 1 Collect Data. Collect data sets, number series or system information that have a relationship or influence to a system failure from relevant sources such as classification societies, ship owners, flag states, insurance companies and experts. 

Step 2 Prepare Data Define and prepare training input sets and decide how to handle the gathered information for presentation to the ANNs. Determine the range of data and set minimum and maximum values to these levels. 

Step 3 Extract test data set hi order to be able to test the trained networic, it is common to set aside some of the data for testing (cross validation). Usually, the total data set is divided into two, one for training and the other for testing. 

Step 4 Train the network Select suitable network architecture by setting the number of network inputs equal to the number of input variables and the number of network outputs equal to that of problem output. Select the number of hidden layers and the number of neurons in each hidden layer. 

Step 5 Test the ANN Apply the test data sets to the trained ANN model to test its 

---

A risk estimation framework incorporating artificial neural networks (ANNs) is described with two case studies demonstrating its use. Some suggestions are made for further research and development on ANN techniques in the context of maritime safety assessment. The possibility to pool records on notation fields such as system data, function information and casualty or defect data in an agreed and standardised database structure is discussed. 

A risk estimation framework incorporating ANNs is depicted in a flowchart as seen in Figure 10.2. The framework comprises the following steps (Sii (2001), Wang et al. (2001)) ... 

---