Professional Documents
Culture Documents
Artificial Neural Network (ANN) Model of Railway Accident Analysis by SPSS
Artificial Neural Network (ANN) Model of Railway Accident Analysis by SPSS
net/publication/336149500
CITATIONS READS
0 189
1 author:
Saurav Barua
Daffodil International University
52 PUBLICATIONS 41 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Effectiveness of Geogrid for Protecting Polder in Sundarban Coastal Region View project
All content following this page was uploaded by Saurav Barua on 01 October 2019.
Independent variables: Accident causes, Accident location, Involvement of vehicle of pedestrian and time
of accident.
Method of modeling: Multi-layer perceptron using Artificial Neural Network
5.Improper maintenance of
the train tracks
5
3 Accident 1.Midblock 1 AL
Location
2.level crossing 2
3.Near station 3
4 Involvement of 1.Large Vehicle (Bus, IVP
vehicle or Truck) 1
pedestrian
2.Car, Micro Bus 2
3.Train 3
4.NMV 4
5.Pedestrian 5
5 Time of 1.Day time 1 TA
Accident
2.Night time 2
[Date] 1
Saurav Barua
Assistant Professor,
Daffodil International University
Variable view:
[Date] 2
Saurav Barua
Assistant Professor,
Daffodil International University
Variable assigning:
[Date] 3
Saurav Barua
Assistant Professor,
Daffodil International University
Partitions:
Architecture:
[Date] 4
Saurav Barua
Assistant Professor,
Daffodil International University
Output parameters:
Then click OK
[Date] 5
Saurav Barua
Assistant Professor,
Daffodil International University
Step 4: Output:
Network Information
Input Layer Covariates 1 AC
2 AL
3 IVP
4 AT
Number of Units a 4
Rescaling Method for Covariates Standardized
Hidden Layer(s) Number of Hidden Layers 1
Number of Units in Hidden Layer 1a 2
Activation Function Hyperbolic
tangent
Output Layer Dependent Variables 1 FT
Number of Units 2
Activation Function Softmax
Error Function Cross-entropy
a. Excluding the bias unit
[Date] 6
Saurav Barua
Assistant Professor,
Daffodil International University
Model Summary
Training Cross Entropy Error 43.190
Percent Incorrect Predictions 8.1%
Stopping Rule Used 1 consecutive
step(s) with no
decrease in errora
Training Time 0:00:00.06
Testing Cross Entropy Error 14.185
Percent Incorrect Predictions 5.3%
Dependent Variable: FT
a. Error computations are based on the testing sample.
[Date] 7
Saurav Barua
Assistant Professor,
Daffodil International University
Parameter Estimates
Predicted
Hidden Layer 1 Output Layer
Predictor H(1:1) H(1:2) [FT=1] [FT=2]
Input Layer (Bias) -2.019 -.152
AC -2.716 -1.062
AL 1.067 -2.155
IVP -.561 .894
AT -.208 1.741
Hidden Layer 1 (Bias) -2.395 1.989
H(1:1) -2.093 2.347
H(1:2) 3.707 -3.225
Classification
Predicted
Sample Observed 1 2 Percent Correct
Training 1 27 17 61.4%
2 0 165 100.0%
Overall Percent 12.9% 87.1% 91.9%
Testing 1 17 5 77.3%
2 0 73 100.0%
Overall Percent 17.9% 82.1% 94.7%
Dependent Variable: FT
Independent Variable
Importance
Normalized
Importance Importance
AC .213 56.0%
AL .380 100.0%
IVP .217 57.1%
AT .191 50.3%
[Date] 8
Saurav Barua
Assistant Professor,
Daffodil International University
[Date] 9