Skip to content

This is a Proof of Concept (POC) project evaluating the output & performance of a Neural Network (NN) for a small-sized insurance company. We want to prove that applying Data Science principles and deep learning models can provide value to all types of businesses, even with limited dataset.

Notifications You must be signed in to change notification settings

aliagowani/Deep_Learning_Customer_Classification

Repository files navigation

Description and Overview

Evaluating Output & Performance of a Neural Network for Small-Sized Insurance Company

This is a Proof of Concept (POC) project evaluating the output & performance of a Neural Network (NN) for a small-sized insurance company. We want to prove that applying Data Science principles and deep learning models can provide value to all types of businesses, even with limited dataset.

We leverage a real-case data from Texas Giant Insurance (TGI). TGI focuses on providing commercial and personal insurance programs to its clients. TGI is an independent insurance company with an in-depth knowledge of multiple insurance products and carriers. They proactively provide many services to their policyholders.

KDE of Customer and their duration

Supervised Learning Models Evaluation

We evaluated three models and showed that the deep learning model is the most accurate, even with the limited features and a small dataset. We will look at accuracy though, it may be argued that F1 Score may be more relevant:

  1. Supervised Learning (Logistic Regression) Model: 73.9% (F1 Score: 78.7%)
  2. Baseline Single-Layered NN: 69.4%
  3. Bayesian Optimized Deep Learning NN: 77.8% (F1 Score: 84.6%)

Bayesian Optimization Search

Confusion Matrix of the Bayesian Optimized Model

We are pleased to report the following findings from phone calls with customers:

  • Two customers who are no longer TGI customers said they would be interested in coming back but are price sensitive.
  • Two customers who are no longer TGI customers are interested in return; though, TGI may prefer not for them to come back.
  • One customer who is still a customer asked for a quote to another line of insurance products (automobile). This was a pleasant surprise and may provide future up-selling opportunities.
  • The client confirmed the robustness of the model that they suggested using this model monthly to identify and prioritize existing customers and potential customers that would return to the company.

Note: The Customer IDs have been modified to preserve customer privacy.

About

This is a Proof of Concept (POC) project evaluating the output & performance of a Neural Network (NN) for a small-sized insurance company. We want to prove that applying Data Science principles and deep learning models can provide value to all types of businesses, even with limited dataset.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published