Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

An Artificial Neural Network

Approach for Road Accident


Severity Prediction
INTRODUCTION
 Determining the underlying causes and severity of road accidents is a major
concern worldwide. Several techniques have been used over the years, but more
recently, Machine Learning (ML) approaches, especially Artificial Neural
Networks (ANN), seems to be crucial.
 This research work focuses on the prediction of accident severity in the island of
Mauritius.
DATA PREPARATION
 The data pre-processing consisted of the removal of non-contributing and
conflicting variables treatment of imbalance class, one-hot encoding of the
categorical variables and min-max normalization of the numerical variables
were carried out to obtain a clean dataset.
METHODOLOGY
 In this work, the performance of the ANN using the sklearn MLP package from Python was
compared using-
➢ 1 hidden layer of 50 neurons
➢ 2 hidden layers of 50 and 25 neurons
➢ 3 hidden layers with 50, 25 and 10 neurons
► Among the three deep layer models (Artificial neural network, Multi-layer perceptron, Cross-
validation) the most performant one was then compared with five machine learning algorithms
namely
► support vector machine,
► gradient boosting,
► logistic regression,
► random forest and
► naïve bayes which were previously assessed in using training and testing ratios of 75:25,
80:20 and 90:10 respectively.
RESULTS

 The stratified 10-fold cross-validation resulted in a mean accuracy of 82.6%


with a standard deviation of 0.01.
 The results of the experimentation of the ANN with 50 and 25 neurons in the
hidden layers respectively showed the highest accuracy of 84.1% using a split
ratio of 90:10.
 An increase in the accuracy of the model when the hidden layer is incremented
from one to two. However, adding a third hidden layer caused the accuracy to
drop
CONCLUSION AND FUTURE WORK

 Since the ANN involves several hyperparameters that affect its performance, in
future work, it is proposed to experiment with other deep learning models to
further enhance the performance for better prediction.

You might also like