Insurance claim prediction - Analytic Vidhya
My primary objective was to make the model rich in features. I created a significant number of features and then used feature selection to narrow it down to the best 25.
I started with 10 fundamental features, but the results were unsatisfactory. Creating new features gave me a good result. I tried out several other classification algorithms, including Random Foreset, GradientBoost, and LGBM, and ultimately settled on GradientBoost. The imbalance was addressed by employing SMOTENN. SMOTE and other methods were also attempted, but SMOTENN proved to be the most effective.
- Position: 76th
- score: 0.1684487580
Due to timing constraints, hyperparameter tweaking was not possible. The final performance is determined by the model's default parameters.
- #1 https://docs.google.com/document/d/14CqefxCVxlZjUOG5QCmhIqpXguVVv7_9Ug3l7CI5Bek/edit
- #2 https://github.com/Santanu7/analytics-vidya-dataverse/blob/main/Dataverse_Claim_Prediction_Sol.ipynb
- #10 https://github.com/sharma-ji/av_dataverse2022
- #11 https://github.com/kc113/Dataverse-Hack
- #12 https://docs.google.com/document/d/1nS5eyYSiv30HlsoCV9s8m3FniUER2kruGbsFYvtelMo/edit#heading=h.215aoooqhg0b
- #17 https://docs.google.com/document/d/1ZAtjMjigWwK8nusE24Xf7n1kH7aont_V/edit