Performance of Multiple Linear Regression and Artificial Neural Network in Predicting Risk Index

Abstract

The act of deliberate damage to oil and gas equipment and facilities
has been a common phenomenon in Nigeria and has posed huge risk
of economic, social and political effects on the people, companies and
the environment. The target of this study is to assess the performance
of multiple linear regression (MLR) and artificial neural network
(ANN) for the prediction of risk index. Twenty-six (26) years
secondary data were obtained from the archive of the Nigerian
National Petroleum Corporation (NNPC), covering the period of 1989
to 2016 and capturing information on risk index, vandalism, rupture,
spillage and volume. To assess the quality of the data, preliminary
analysis involving; reliability analysis using one-way analysis of
variance and test of homogeneity was done. To examine the
correlation and determine the exact relationship between the risk
index and other independent variables, selected linear and non-linear
models were employed. For the linear model, multiple linear
regression (MLR) was used and for the non-linear model, artificial
neural network (ANN) was employed. The calculated p-value of 0.000
based on the reliability test shows that the data are reliable and
adequate. On the overriding influence of the selected independent
variables, vandalism was observed to be positively correlated with
risk index thus making this variable the most significant variable that
influence risk index compares to other independent variables such as
rupture, spills and volumes. On the best fit model for risk index
prediction, result of regression plot of output revealed that; artificial
neural network with R2
value of 0.9973 was acclaimed the best model
ahead of multiple linear regression with R2
value of 0.5640.