An explanation of the entirety of the competition can be found in this Medium Article - Medium Read Link
This case study is based on the Kaggle Competition: https://www.kaggle.com/c/mercari-price-suggestion-challenge
Notebook Contents -
| S.No | Section | Jupyter Notebook |
|---|---|---|
| 1. | Exploratory Data Analysis | EDA-CS1.ipynb |
| 2. | Data Preprocessing & Feature Engineering | Feature_Engg_v6.ipynb |
| 3. | Data Preprocessing & Feature Engineering for Test data | Feature_Engg_TestData_v6.ipynb |
| 4. | Test Model 1: Lasso | BaselineModelFinal_v4_BEST(BN+RELU).ipynb |
| 6. | Test Model 2: Ridge | BaselineModelFinal_v4_BEST(BN+RELU).ipynb |
| 7. | Test Model 3: ElasticNet | BaselineModelFinal_v4_BEST(BN+RELU).ipynb |
| 8. | Test Model 4: SVR | BaselineModelFinal_v4_BEST(BN+RELU).ipynb |
| 9. | Test Model 5: XGBoost | BaselineModelFinal_v4_BEST(BN+RELU).ipynb |
| 10. | Test Model 6: LGBM | BaselineModelFinal_v4_BEST(BN+RELU).ipynb |
| 10. | Test Model 7: MLP-256 with 2layers | BaselineModelFinal_v4_BEST(BN+RELU).ipynb |
| 10. | Test Model 8: MLP-1024 with 6layers | BaselineModelFinal_v4_BEST(BN+RELU).ipynb |
| 11. | Final Ensemble Model of MLP-256 and MLP-1024 | BaselineModelFinal_v4_BEST(BN+RELU).ipynb |
| 12. | Final Pipeline Function 1 - Prediction Output | Final_Submission_Case_Study1.ipynb |
| 13. | Final Pipeline Function 2 - RMSLE Metric Output | Final_Submission_Case_Study1.ipynb |
Final Loss Table -
| Model | Features used | CV Loss = RMSLE |
|---|---|---|
| Lasso Regression | TF-IDF + Feature Combined for name and text | 0.745 |
| Ridge Regression | TF-IDF + Feature Combined for name and text | 0.444 |
| Elastic Net | TF-IDF + Feature Combined for name and text | 0.745 |
| SVR | TF-IDF + Feature Combined for name and text | 0.687 |
| XGBoost | TF-IDF + Feature Combined for name and text | 0.450 |
| LGBM | TF-IDF + Feature Combined for name and text | 0.439 |
| MLP Model - 256 | TF-IDF + Feature Combined for name and text | 0.420 |
| MLP Model - 1024 | TF-IDF + Feature Combined for name and text | 0.412 |
| Ensemble Model (MLP-256 + MLP-1024) | TF-IDF + Feature Combined for name and text | 0.4047 |