Predictive model for insurance agency data.
Agency data intuitively has value to an agency and potentially to agent's carrier partner(s). One way to gain value from that data - which exists within agency management systems - is to create one or more predictive models. Gaining insight from predictive modeling using agency data is the fundamental nature of this effort.
For the R code, there are a bunch of files, but the main (working) files are naive_bayes_model.rmd and knn_model.rmd. Those are the R Markdown files with narration. The other R files are works in progress or just .r versions of the .Rmd files.
For the Python code, the main file is dt_classifier_tuned.ipynb, which is the Jupyter Notebook. The other files are the base .py files and can be used for reference.
This entire project was completed using RStudio RStudio and Python (Jupyter Notebook). You'll need R, an R IDE (if you prefer), and a Python distribution that includes Jupyter Notebook. I used RStudio and Anaconda (with PyCharm).
You'll need the following R packages installed using install packages("PACKAGE_NAME"):
- xlsx
- dplyr
- e1071
- caret
- formattable
There should be no issue installing these packages if you have RStudio.
You'll need the following Python libraries installed using (pip install <PACKAGE_NAME>):
- pandas
- numpy
- sklearn
It's clear that predictive modelling with agency has value. The contents of this analysis are a single perspective of such an effort, although subsequently, at least one other perspective is presented to give the reader a glimpse into how varied this analysis could potentially be.
This uses a largely unalered simple Naive Bayes model to set a baseline for subsequent predictive models, which can be created using other algorithms.