How to use default model:

• Upload crash data or narratives in the form of a csv file.


• Choose from default classification algorithms and run models for crash classification.


• Provide an email id and a 3-digit pin to run the models.This is required for the tool to distinguish data from multiple users and store individual classification results in the database.

Default Model Page

Noisy-OR Classification:

• To apply Noisy-OR classifier to crash narratives, at first the word probabilities of unigram and bigram are computed.


• Then by combining these probabilities using the Noisy-OR method the final classification score is calculated.


• Equation 1 and 2 below shows the calculation process of word probability and Noisy-OR score respectively.

Noisy-OR Equations

Explanation of text colors in Tabular results:

• Orange color represents Unigram.


• Green color represents Bigram.


• Green-orange color represents overlaping words of unigrams and bigrams.


• Black color words are considered to be less significant for clasification.


• Color intensity represents the weight of the words while classifying the text/reports. A minimum threshold value of 0.35 has been used.


• 'Click' on a result will display its location on Map.

Result Page

Sample result:

unit 1 and unit 2 were inbound on e. washington ave. prior to ingersoll st. unit 1 was in the left-most lane closest to the median. unit 2 was in the next lane to the right. the driver of unit 1 had a medical event and lost consciousness while driving. unit 1 struck a road maintenance lighted directional sign and traffic cones. the passenger of unit 1 engaged the emergency brake to bring the vehicle to a stop. unit 2 reports being struck by the flying debris from the crash. unit 1 sustained front end damage. unit 2 reported scratches and damage to the front, driver-side corner of the vehicle and the driver-side rear door. the maintenance sign also sustained damage to the light panel and the body of the sign. no citation issued. lt. fiore and sgt. doberstein on scene.

How to train model on the go:

• Upload training data in the form of a csv file.


• Choose a model type to train.


• Provide an email ID and a 3-digit pin to run the models.This is required for the tool to distinguish data from multiple users and store individual classification results in the database.

Train Model Page

Noisy-OR Classification:

• To apply Noisy-OR classifier to crash narratives, at first the word probabilities of unigram and bigram are computed.


• Then by combining these probabilities using the Noisy-OR method the final classification score is calculated.


• Equation 1 and 2 below shows the calculation process of word probability and Noisy-OR score respectively.

Noisy-OR Equations

Trained Model Summary:

• Get an idea of training from summary statistics of word probability.


• Choose best cut off value from table of evaluation metrics for crash classification. Use the best cut-off value of word probability to the next step.


• Scroll down to the bottom of the page and proceed to use model upon satisfactory training summary.

Training Summary Page

How to use trained model:

• Upload Test data in csv file format.


• Use the best cut off value from the previous table of evaluation metrics for training.


• Classify the text

Test Model Page

Explanation of text colors in Tabular results:

• Orange color represents Unigram.


• Green color represents Bigram.


• Green-orange color represents overlaping words of unigrams and bigrams.


• Black color words are considered to be less significant for clasification.


• Color intensity represents the weight of the words while classifying the text/reports. A minimum threshold value of 0.35 has been used.


• 'Click' on a result will display its location on Map.

Result Page

Sample result:

unit 1 and unit 2 were inbound on e. washington ave. prior to ingersoll st. unit 1 was in the left-most lane closest to the median. unit 2 was in the next lane to the right. the driver of unit 1 had a medical event and lost consciousness while driving. unit 1 struck a road maintenance lighted directional sign and traffic cones. the passenger of unit 1 engaged the emergency brake to bring the vehicle to a stop. unit 2 reports being struck by the flying debris from the crash. unit 1 sustained front end damage. unit 2 reported scratches and damage to the front, driver-side corner of the vehicle and the driver-side rear door. the maintenance sign also sustained damage to the light panel and the body of the sign. no citation issued. lt. fiore and sgt. doberstein on scene.