Analysis Of Impact Of Weather Related Factors On Road Crashes Using Data Mining Techniques

Road traffic safety is the result of a complex collaboration of factors,and the roots behind road vehicle crashes require different measures to reduce their impacts.This study evaluated how strongly the variation in crash severity levels is associated with weather conditions.As because of rapid urbanization,motorization has also expanded in developed countries leading to an increase in road traffic accidents.Weather impacts on freeway traffic safety have become a growing concern for roadway safety agencies.The majority of most weather-related crashes happen on wet pavement and during rainfall.Therefore,it is very important to explore the relationship between weather related variables and road accidents severity by using data mining techniques approach.Although,previous literature is present on weather impact on road crashes but impact of variation in each weather factor on crash severity need to be explained in detail for traffic safety purpose.This study is performed to analyze and quantify the impact of weather related factors(National Climatic Data Center data)on road traffic accident severity,based on Highway Safety Information System(HSIS)accident dataset.In order to find a better model fitted related variables,four candidate models order logit(OL),Decision Tree,Random Forest and neural networks were chosen to examine in Stata and python respectively.During the model construction,seven major factors related to weather condition are considered.They are air temperature,average wind speed,daily rain,monthly rain,yearly rain,wind gust and relative humidity.After fitting statistical and machine learning models sensitivity analysis is also performed to deeply analyze the effect of weather factors on accident injury severity from machine learning models performances.The results of sensitivity analysis showed that the variation of different weather related parameters have different impact on machine learning models based on accident severity data.The models results showed that MLP performance by accuracy is higher than RF and Decision tree.The MLP.RF and DT models showed the predictive accuracy for fatal injury severity by 82.04%,81.18% and 80.58% respectively.The findings of the study also helped in identifying areas in policy and education that need improvement.