Author

Problem Statement :

Task 1 :

Create a predictive model which will help the insurance marketing team to know which customer will claim the insurance..

Task 2 :

Suggestions to the Insurance market team to make customers buy the product.

Here are some of the key outcomes of the project:

• The Dataset was large, totally around 6 lakhs samples & before preprocessing 2% of the datasamples were dropped.
• The samples were highly imbalanced hence over sampling, undersampling, SMOTE Technique were applied on the data to balance the classes to the dataset.
• Visualising the distribution of data & their relationships, helped us to get some insights on the relationship between the feature set.
• Feature Selection/Eliminination was carried out and appropriate features were shortlisted.
• Testing multiple algorithms with fine-tuning hyperparamters gave us some understanding on the model performance for various algorithms on this specific dataset.
• The boosting & ensemble algorithms perform the best on the current dataset.

Problems :

• The data is non linear and highly imbalanced which resulted in Over-fitting.
• It was difficult to understand data and pre-process it due to the data and feature names being irrelevant.
• The run time was high during fitting the model.
• Accuracy was good for few models but the problem was presence of several outliers whereas for few models accuracy was poor.
• There are misclassifications and handling them was difficult.
• GridSearchCV and RandomizedCV consumed lot of time and increase in parameters gave memory error.

Model Comparision Report :

• We used Logistic Regression, Decision Tree, Random Forest, XGBoost along with OverSampling, UnderSampling also SMOTE and PCA.
• The final model we applied is SEQUENTIAL MODEL which is preferred among all the models as it gave the best accuracy.