Geoscience Reference
In-Depth Information
Table 3
The output factor
used in the study
Risk score
Risk category
Less than 4.400
Very low
4.400-4.600
Low
4.600-5.600
Medium
5.600-6.600
High
More than 6.600
Very high
Fig. 12
Architecture of an
MLP neural network
To calculate the risk score of each record we have used cumulative sum of 16
factors as given below.
Total Risk
= (
Weight
1
×
Risk Value
1
) + (
Weight
2
×
Risk Value
2
)
+ ......+ (
Weight
16
×
Risk Value
16
)
As we need to submit risk scores to the neural network to make it learn from sam-
ples we need to formalize output data for each record in universal set. In the study,
the output factor (i.e., risk score) has been converted into five-level categorical
variable ranging from “very high” to “very low” as given in Table
3
.
4.2 Training and Testing of Proposed Neural Network
In the study we have used Multilayer Perceptron (MLP) as a neural network struc-
ture. MLP is one of the most frequently used neural network architectures in both
classification and prediction purposes and it belongs to the class of supervised
neural networks (Fig.
12
).
In our training process we have used 30.000 randomly selected records from
the universal set. We have developed a software to maintain works with neural net-
work models using Java programming language. Our all data used for training and
testing is stored in Oracle database.
