Case Sensitivity Analysis Factors in Traffic Accidents
Content
Chapter Notes: • • Neural Network = brain metaphor for information processing Ability o to “learn” from the data o non parametric nature (no rigid assumptions) o ability to generalize Neural computing = patter‐recognition methodology for machine learning Artificial Neural Network = resulting model from neural computing NN computing key component of any data mining tool kit ANN posses these desirable traits o Learn o Self‐organize o Support tolerance Black‐box testing is the primary approach for verifying that inputs produce the appropriate outputs Sensitivity analysis is front runner of the techniques proposed for shedding light into the “black‐box” characterization of trained neural networks Sensitivity analysis is a method for extracting the cause‐and‐effect relationships among inputs and outputs of a trained NN model
• • • •
• • • •
Case Notes: • • • • • According to National Highway Traffic Safety (NHTSA) over 6 million traffic accidents claim more than 41,000 lives each year in US Traffic‐safety researchers have an special interest in causes of accident and related injury severity Research goal is not only reducing the number of accidents but also the severity of injury To accomplish the reducing of severity in injury is to identify the most profound factors that affect it Understanding the circumstances under which the drivers and passengers are more likely to be severely injured (or killed) in a car accident can help to improve the overall driving safety situation Factors that potentially increase the risk of injury severity are: o Demographic: age, gender o Behavioral: seatbelt usage, usage of drugs or alcohol while driving o Environmental : roadway conditions, surface conditions, weather, light conditions, direction of the impact, vehicle orientation in the crash, occurrence of a rollover o Vehicle technical aspects: vehicle`s age, body type Exploratory data mining study ‐to identify which factors became increasingly more important‐
•
•
•
•
•
•
•
used a large data sample: 30,358 police‐reported accident records obtained from the general estimates of NHTSA o included accidents geographically representative sample of multiple vehicle collision accidents, single‐vehicle fixed‐object collision and single‐vehicle noncolision (rollover) crashes Many previous studies o had primarily used regression‐type generalized linear models o function relationship between injury severity and crashed related factors are assumed to be linear Artificial Neural Network (ANN)are known to be superior in capturing highly nonlinear complex relationships between the predictor variables (crash factors) and the target variable (severity level of the injuries), therefore they decided to used a series of ANN models to estimate the significance of the crash factor on the level of injury severity sustained by the driver From methodological standpoint, they followed a two‐step process o First: develop a series of prediction models (one for each severity injury level) to capture the in‐depth relationships between the crash‐related factors and a specific level of injury severity o Second: conducted a sensitivity analysis on the trained neural model to identify the prioritized importance of crash‐related factors as they relate to different injury severity levels In the formulation of the study the five‐class prediction problem was decomposed into a number of binary classification models in order to obtain granularity of information needed to identify the “true” crash‐related factors and different levels of injury severity Results revealed: o Considerable differences among the models built for different injury severity levels. This implies that the most influential factors in prediction models highly depend on the level of injury severity o The seatbelt usage was the most important determinant variable for predicting higher levels of injury severity but it was one of the least significant predictors for lower levels of injury severity o Driver’s Gender was among the significant predictors for lower levels of injury severity but it was not among the significant factors for higher levels of injury severity o Age becomes an increasingly more significant factor as the level of injury severity increases o
• • • •
• • • •
Case Notes: • • • • • According to National Highway Traffic Safety (NHTSA) over 6 million traffic accidents claim more than 41,000 lives each year in US Traffic‐safety researchers have an special interest in causes of accident and related injury severity Research goal is not only reducing the number of accidents but also the severity of injury To accomplish the reducing of severity in injury is to identify the most profound factors that affect it Understanding the circumstances under which the drivers and passengers are more likely to be severely injured (or killed) in a car accident can help to improve the overall driving safety situation Factors that potentially increase the risk of injury severity are: o Demographic: age, gender o Behavioral: seatbelt usage, usage of drugs or alcohol while driving o Environmental : roadway conditions, surface conditions, weather, light conditions, direction of the impact, vehicle orientation in the crash, occurrence of a rollover o Vehicle technical aspects: vehicle`s age, body type Exploratory data mining study ‐to identify which factors became increasingly more important‐
•
•
•
•
•
•
•
used a large data sample: 30,358 police‐reported accident records obtained from the general estimates of NHTSA o included accidents geographically representative sample of multiple vehicle collision accidents, single‐vehicle fixed‐object collision and single‐vehicle noncolision (rollover) crashes Many previous studies o had primarily used regression‐type generalized linear models o function relationship between injury severity and crashed related factors are assumed to be linear Artificial Neural Network (ANN)are known to be superior in capturing highly nonlinear complex relationships between the predictor variables (crash factors) and the target variable (severity level of the injuries), therefore they decided to used a series of ANN models to estimate the significance of the crash factor on the level of injury severity sustained by the driver From methodological standpoint, they followed a two‐step process o First: develop a series of prediction models (one for each severity injury level) to capture the in‐depth relationships between the crash‐related factors and a specific level of injury severity o Second: conducted a sensitivity analysis on the trained neural model to identify the prioritized importance of crash‐related factors as they relate to different injury severity levels In the formulation of the study the five‐class prediction problem was decomposed into a number of binary classification models in order to obtain granularity of information needed to identify the “true” crash‐related factors and different levels of injury severity Results revealed: o Considerable differences among the models built for different injury severity levels. This implies that the most influential factors in prediction models highly depend on the level of injury severity o The seatbelt usage was the most important determinant variable for predicting higher levels of injury severity but it was one of the least significant predictors for lower levels of injury severity o Driver’s Gender was among the significant predictors for lower levels of injury severity but it was not among the significant factors for higher levels of injury severity o Age becomes an increasingly more significant factor as the level of injury severity increases o
