Road_Severity_Accident_Prediction_WebApp

Road Severity Accident Prediction Using Random Forest Classifier!

Check Out The Web App at: road_accident_severity.io

Problem Statement

The task is to classify this variable based on the other 31 features step-by-step by going through each day's task.

• The target feature is Accident_severity which is a multi-class variable.
• The metric for evaluation will be f1-score

Dataset Used

Source of the dataset: Click Here

• The data set is collected from Addis Ababa Sub-city police departments for master's research work.
• The data set has been prepared from manual records of road traffic accidents of the year 2017-20.
• All the sensitive information has beenexcluded during data encoding and finally it has 32 features and 12316 instances of the accident.
• Then it is preprocessed for identification of major causes of the accident by analyzing it using different machine learning classification algorithms.

Dataset Info

The following are metadata regarding the dataset:

• Time : Time of the accident (in 24 hours format) |
• Day_of_week : Day of the week when the accident occurred
• Age_band_of_driver :The age group of the driver
• Sex_of_driver : Gender of driver
• Educational_level : Driver’s highest education level
• Vehical_driver_relation : What’s the relation of a driver with the vehicle
• Driving_experience : Number of years of driving experience the driver has
• Type_of_vehicle : What’s the type of vehicle
• Owner_of_vehicle : Who’s the owner of the vehicle
• Service_year_of_vehicle : The last service year of the vehicle
• Defect_of_vehicle : Is there any defect on the vehicle or not?
• Area_accident_occured : Locality of an accident site
• Lanes_or_Medians : Are there any lanes or medians at the accident site?
• Road_allignment : Road alignment with the terrain of the land
• Types_of_junction : Type of junction at the accident site
• Road_surface_type : A surface type of road
• Road_surface_conditions : What was the condition of the road surface?
• Light_conditions : Lighting conditions at the site
• Weather_conditions : Weather conditions
• Type_of_collision : What is the type of collision
• Number_of_vehicles_involved : Total number of vehicles involved in an accident
• Number_of_casualties : Total number of casualties
• Vehicle_movement : How the vehicle was moving before the accident occurred
• Casualty_class : A person who got killed during an accident
• Sex_of_casualty : What the gender of a person who got killed
• Age_band_of_casualty : Age group of casualty
• Casualty_severtiy : How severely the casualty was injured
• Work_of_casualty : What was the work of the casualty
• Fitness_of_casualty : Fitness level of casualty
• Pedestrain_movement : Was there any pedestrian movement on the road?
• Cause_of_accident : What was the cause of an accident?
• Accident_severity : How severe an accident was? (Target variable)

Steps Involved

All steps involved - Can be found in notebook too!

EDA

• Importing the Dataset
• Check Metadata of the Dataset
• Find Number of Missing Values Present in Each Column
• Target Variable Classes Distribution & Visualization
• Check Education Levels of Drivers
• Auto Data Visualization Using the Dabl Library
• Check for Association Between Road Surface Type and Accident Severity

Data Preprocessing & Feature Engineering

• Converting the “Time” Column
• Feature Selection General
• Missing Value Treatment
• One-Hot Encoding Using get_dummies()
• Target Encoding using LabelEncoder()
• Feature Selection Using Chi2 Statistic
• Imbalanced Dataset Treatment Using SMOTENC Technique

Model Building, Evalaution & Saving

• Building a Random Forest Classification Model
• Evaluation: Classification Report & F1 Score
• Saving the Model & Encoder Object using joblib

Deploying

• Loading Model & Encoder
• Preparing Streamlit Project Requirements
  • requirements.txt
  • github repo
• Creating app.py file which includes:
  • Initialising & Importing
  • User Input & Model Prediction
  • Webpage Aesthetics
• Deploying on Streamlit Community Cloud

DONE!

---

Reach out to me on: Twitter | Linkedin | Blog(AV)