Inspiration
Building real world projects provides conceptual and practical clarity and it is beneficial for the society.
What it does
This project is built as a model trained on an accident dataset of India (can be trained on any accident dataset), it predicts the type, severity and reason of accident.
How we built it
Using Python and it's useful machine learning package "Scikit-Learn", with other supporting packages for mathematical computation and data visualization.
Challenges we ran into
I used ensemble method algorithm in the code for model training, Gradient Boosting to be exact. It is a complex algorithm which took a lot of runtime. I also had to perform hyper parameter tuning to use the accurate parameters using Grid Search, otherwise the accuracy could have been sacrificed.
Accomplishments that we're proud of
Prior Knowledge of Data cleaning methods came very handy. Overall, my model predicts the outcomes with a whopping accuracy of ~96 percent.
What we learned
Knowing which algorithm is suitable to be applied on which dataset plays an important part in defining the accuracy of model, with other factors like Feature Engineering, Data Encoding, Scaling and Hyperparameter Tuning., division of training and testing dataset and many more!
What's next for Traffic Accidents Analysis
I look forward to create more real world models using time series data, using neural networks etc. this time, a chatbot, other related stuff using Tensorflow, keras etc, frameworks. I would love to use what I will learn forward in creating a fully fledged chatbot model for such real world projects so that common public can use them.
Built With
- matplotlib
- numpy
- pandas
- python
- scikit-learn
- seaborn
- vscode
