his is a multiclass classification project to classify the severity of road accidents into three categories. The project is based on real-world data, and the dataset is highly imbalanced. The goal of this project is to identify major causes of accidents and classify the severity of accidents.
The data set is collected from Addis Ababa Sub-city, Ethiopia police departments for a master's research work. The data set has been prepared from manual records of road traffic accidents of the year 2017-20. The dataset has 32 features and 12,316 instances of the accident. The target feature is "Accident_severity", which is a multi-class variable.
The streamlit app is used for deploying the model.
-
Exploratory data analysis
- Data analysis using dabl
- Exploratory data analysis using matplotlib and seaborn
-
Data preparation and pre-processing
- Missing Values Treatment using fillna method
- One Hot encoding using pandas get_dummies
- Feature selection using chi2 statistic and SelectKBest method
- PCA to reduce dimensionality
- Imbalance data treatment using SMOTENC technique
-
Modelling using sci-kit learn library
- Baseline model using RandomForest using default technique
- Tuned hyperparameters using n_estimators and max_depth parameters
-
Evaluation
- Evaluation metric was weighted f1_score
- Baseline model evaluation f1_score = 61%
- Final model evaluation f1_score = 88%
To get started with the project, you can simply clone the repository using the following command:
git clone https://github.com/yourusername/Road-Traffic-Severity-Classification.git
After cloning the repository, install the required packages using the following command:
pip install -r requirements.txt
Run the streamlit application:
streamlit run app.py
The project was able to identify the major causes of accidents and classify the severity of accidents. The final model was able to achieve an f1-score of 88%, which is a significant improvement over the baseline model. This project can be useful for policymakers and stakeholders to identify major causes of accidents and take necessary actions to reduce accidents.