Car Crash Severity for Autonomous Driving Capstone Project

Capstone Project for Data Science Course

Overview

Detection of car crash severity with the background:

• Growth interest in AI systems, especially Autonomous Driving
• Prediction by governments to automate driving systems (BCG, 2015)
• Safety is still intricate issue, therefore:
  • Observe variables that took part in car crash severity from real-life data
  • Comparison of best machine learning methods to classify severity
• Model is also attractive to build an alert system for city infrastructure workers (paramedics, police, firefighter, etc)

Data

• Seattle GeoData. The data is an open data from the Seattle government, collected from 2004-2020. (Collisions_OD).
• Data specifications: 40 attributes, 221,144 collection of accidents. Severity indicator:
  • 3: Fatality — High Probability
  • 2b: Serious Injury — Mild Probability
  • 2: Injury — Low Probability
  • 1: Property Damage — Very Low Probability
  • 0: Unknown — Little to No Probability

Requirements

• Language : Python

Libraries:

• matplotlib :3.2.1
• pandas : 1.0.5
• scikit-learn : 0.22.2
• numpy : 1.18.5
• jupyter notebook : 5.2.2

Description

Preprocessing Steps

• Data cleaning
• Fill NaN values
• Variable correlation

Model Architecture

Three models are evaluated:

• K-Nearest Neighbor
• Decision Tree
• Logistic Regression All models are previously searched in the space of [1,20] for k, [1,15] for depth and [0.001,0.01,0.1,1,10,100] for regression in logistic regression.

Evaluation

Evaluation is done using 5 metrics: Jaccard-index, F1-score, LogLoss, Precision, Recall