This is a submission project for the petfinder.my kaggle challenge:
https://www.kaggle.com/c/petfinder-adoption-prediction
| Idea | Description | Status |
|---|---|---|
| Democratic Ensembles | When ensembling the models you can go for multiple strategies. For Example: democratic, uniform, median | Not implemented |
| Optimized Rounding | Classic roundings use .5 as threshold. But maybe when rounding the classification with kappa as scorer another threshold is needed | Implemented |
| Regression | This problem could also be solved using a regression model instead of classification. In combination with optimized rounding it could improve the accuracy. Maybe even ensemble the different types after | Not implemented |
| Feature | Description | Fields to use | Status |
|---|---|---|---|
| Name/NoName | Maybe there is a relation between pets who already have names and pets who don't have a name | Name | Implemented |
| Name length | Does the length of a name influence the decision to adopt? | Name | Implemented |
| Cute Names | With online naming databases as a basis, can we categorize names as "cute" or rather "more adoptable"? | Name | Not Implemented |
| Image quality | Maybe there is a relation between the quality of the images and the adoption time | ??? | Not Implemented |
| Length of description | Does the amount of information about the pet influence the decsision to adopt? | calculated field | Implemented |
| sentiment of description | Does the sentiment of the description influence the decision to adopt? | ??? | Implemented |