Abstract:
The epidemic of road accidents growing rate, excides the need for drastic evoke in road safety
within world. The awareness for road accidents has increased within the government and general public, the epidemic of road accidents are increasing an alarming rate. Specific instances
which cause social and economic loss for a particular country can be considered as road accidents. Road safety is the most crucial and vital part of a country. The tragedy of the road
accident is it mainly affects the developing countries by causing huge human losses and negative emotional effects for people. Therefore to establish an accident prevention, reduction
strategies and future forecasting of road accidents with time is a compulsory need. In order
to reduce the severity of the road accident and to enhance road safety, a comparative analysis
of road accident is performed by using machine learning techniques. The main objective of
this study is to identify the most suitable machine learning technique to forecast future accident
counts. In this comparative study forecasting capabilities of CART Regression tree, Multi-Layer
Perceptron (MLP), Support Vector Regression (SVR), Long-Short Term Memory (LSTM) and
Autoregressive Integrated Moving Average (ARIMA) were studied and compared. The comparison was conducted by studying efficiency of algorithms and forecasting of road accidents
data. The results revealed that Long-Short Term Memory (LSTM) provides forecasts which are
similar to the actual predictions with a Root Mean Squad Error (RMSE) of 146.23 while other
techniques only were able to achieve higher RMSE. Multi-Layer Perceptron (MLP) was able
to achieve RMSE of 330.38 which is not effective as the LSTM. Although several researchers
were focused on comparing ARIMA and the Neural Networks, less number of studies focus on
comparing several techniques in machine learning. In this study researcher is focus on forecasting road accident rate by using different machine learning techniques. The identified LSTM
approach will provide effective results when using it as a road accident forecasting model. The
study will provide the necessary information to guide the relevant decision-makers in adopting
suitable measures to reduce the accident rate.
