Fare Prediction
The transportation industry is thriving in the hotspot city. Such companies are providing ridesharing services from door to door using algorithms or chunks of data to leverage the customers at a valuable price. Generally, those fares are dynamic and estimated considering the distance but the target fare should not be dependent only on trip distance. It must consider significant features like the demands, traffic, and potential area based on data from past years. So, the approach considers the service provider and customer to get a relative benefit over the competition. Likewise, the proposed method will include end-to-end development for real-world usage.
Evaluation and Conclusion
| Model Name | RMSE | R2 Score |
|---|---|---|
| Linear Regression | 6.64 | 54.94% |
| Decision Tree | 5.20 | 71.86% |
| Random Forest | 5.26 | 71.49% |
| Gradient Boosting | 4.80 | 76.22% |
The purpose of this project is to estimate those fare calculations, with a machine learning approach that could be fast enough to utilize while providing services to the real-time customer. The results of the regression showed that the proposed method has potential as far as different known quality metrics. Furthermore, I implemented feature scaling and the performance improved. In the deployment, the Gradient Boosting model performed well with 4.47 RMSE & R2 score of 79.22%. The model has been registered in the MLflow so it can be served in real-time as a REST endpoint which allows HTTPS requests and gives the output based on the input request by the client. Later models can be more robust and precise by utilizing real-time traffic data. To sum up, test evaluation and outright insights can be implemented in the food delivery chain or other product delivery services strategically as per the requirement and domain expertise.
Kindly read the project documentation file for getting started. For more details of the project, refer the detailed report file. To download the dataset here.
Copyrights © 2022 by Neel Patel. All rights are reserved.