Modeling
Modeling
- What are the Common price amount always happening for the given list of variables(x-var)?
- This is useful to get a intiution
- This can be used in final stage of modeling (we can tweak the model output , to get desired output at the last)
- Feature Selection using trained models
- Build Different Algorithms to predict the price
- Linear, Lasso,Ridge
- Xgboost,Catboost,LightGBM
- Pytorch model (if needed)
- Baseline model performance metrics
- Measure the model performance using
- Mean Absolute Error
- Mean Squared Error
- RMSE
-
Benchmark Leaderboard of models using Different Hyperparameters
- For above metrics(Error function in the test set)
- Also for Prediction Time
-
Confidence Interval Calculation
- Random Forest Confidence Intervals (using different Decision Trees)
- Gradient Boosting using Quantile Intervals
- Explanation for the prediction
- Shap
- Lime
Developing Baseline Model
we trained 7 Models with no hyperparameter tuning as our first baseline:
- Linear Regression
- Lasso Regression
- Ridge Regression
- RandomForest Regression
- GradientBoosting Regression (GBR)
- CatBoost GBR
- lightGBM GBR
Baseline Score of the trained models
| linear_reg | lasso | ridge | rf | gbr | cat_gbr | light_gbr | |
|---|---|---|---|---|---|---|---|
| maae | 23.9081 | 24.9784 | 23.9086 | 22.1 | 23.3951 | 22.6081 | 22.4571 |
| mae | 52.7526 | 52.9087 | 52.7526 | 55.2584 | 49.892 | 49.4214 | 49.6741 |
| mse | 21382.3 | 21457.7 | 21382.4 | 23178.5 | 20087.1 | 19066.8 | 19289.1 |
| rmse | 146.227 | 146.484 | 146.227 | 152.245 | 141.729 | 138.083 | 138.885 |
- Among all the models catboost Regressor performed well on the test dataset
As we took Catboost model for best prediction accuracy and low-latency predictions and started tuning the hyperparamater of the models
The Training notebook of the catboost model was attached as html in the following link