1 of 12

Adult Mortality Rate

Nang Kaung Shan Kham

Hpu Hpu Thant Sin

Nyan Lin Htut

Kyaw Zaww Linn

2 of 12

Business Scenario

insurance

COMPANY

Project Launch: Initiating a data mining project in the international health insurance sector.
Data Used: Utilizes "Adult Mortality Rate (2019-2021)" dataset.
Clustering Techniques: Employing advanced methods to segment markets based on health, economic, and demographic factors.
Purpose: To create tailored insurance products that align with regional needs.
Business Outcomes: Enhances risk assessment, improves market segmentation, and develops competitively priced products.
Impact: Aims to increase customer satisfaction and drive business growth.

Our company is starting a detailed data mining project to analyze the international health insurance market using a dataset that includes "Adult Mortality Rate (2019-2021)." We plan to use advanced methods to group countries by health data, economic levels, and other factors like GDP per capita and health spending. This will help us create insurance products that match the specific health and economic situations of different regions. This approach will improve how we assess risks, segment the market, and develop insurance products that are priced right and meet the varied needs of people worldwide, which will increase customer satisfaction and help our business grow. Our project utilizes Health and Risk Profile Clustering along with Development-based Clustering to effectively segment the international health insurance market and improve strategic decision-making.

3 of 12

Data processing

Adjusted incorrect data and missing values; minimal initial errors.
Removed outliers like Luxembourg, Ethiopia, and Qatar due to unique profiles.
Categorized GDP and health expenditure into 'low', 'medium', and 'high'.
Skipped normalization/standardization to maintain data scale.
Converted numeric data to categorical for simpler analysis and clearer algorithm results.

Data cleaning was minimal as the dataset was initially clean, focusing on removing incorrect data and handling missing values where necessary. Specific outliers, including Luxembourg, Ethiopia, and Qatar, were excluded due to their unique economic and demographic traits that could distort analyses. The data preprocessing included targeted discretization of GDP per capita and health expenditure into 'low', 'medium', and 'high' categories to simplify economic analysis relevant to insurance and health services. Normalization and standardization were deliberately omitted to maintain the original scale of data, deemed crucial for the business context. Numeric attributes were converted to normal (categorical) attributes to simplify the dataset and improve the interpretability of classification algorithms by reducing variability.

4 of 12

Clustering

Elbow Method Analysis:

Cluster Distinctiveness:

Operational Manageability and Focus:

Chose two clusters based on Elbow Method results.
Additional clusters didn't significantly improve performance.
Two clusters prevent overfitting and capture essential data structure.

Effectively differentiates regions by health expenditures, GDP per capita, and AMR.
Each cluster represents unique economic and health dynamics.
Enables strategic analysis of regional differences.

Balances detailed analysis with operational manageability.
Avoids complexity and ensures statistically significant cluster sizes.
Clusters are focused enough for targeted policy and resource allocation.

Elbow Method Analysis: The decision to use three clusters was based on the Elbow Method, which showed that additional clusters beyond three did not significantly enhance the clustering performance, indicating that three clusters sufficiently captured the essential data structure without overfitting.

Cluster Distinctiveness: The three-cluster solution effectively differentiated regions based on health expenditures, GDP per capita, and Adult Mortality Rates (AMR), each cluster representing unique economic and health dynamics, allowing for a detailed and strategic analysis of regional differences.

Operational Manageability and Focus: Choosing three clusters balanced detailed analysis with operational manageability, avoiding the complexity that more clusters might introduce and ensuring each cluster was large enough to be statistically significant yet focused enough for targeted policy and resource allocation.

5 of 12

Clustering

Data-Driven Decision Making:

Supporting Strategic Planning:

Tailors interventions and policies to the specific needs of each cluster.
Enhances effectiveness and efficiency of healthcare strategies.

Simplifies complex health data.
Supports informed decision-making and strategic planning.
Provides a framework for addressing health and economic challenges regionally.

6 of 12

Health and Risk Profile Clustering

7 of 12

Development-based clustering

8 of 12

Classification

One R

Naive Bayes applied to categorical data for efficient analysis

IBk model used to capture complex patterns among variables

Aimed for detailed mortality risk insights to guide insurance product customization

Utilized Decision Trees for visual segmentation based on economic and health indicators

for straightforward stratification into mortality risk categories

The application of cross-validation across various classification models has reinforced the reliability and stability of our insights into customer behaviors, essential for crafting personalized marketing strategies. The OneR model demonstrated high accuracy in our dataset, though it suggested potential overfitting with its high error rates. The J48 Decision Tree was particularly effective at identifying 'Consumption' as a critical determinant in predicting customer reactions to marketing initiatives, supported by its robust ROC performance. K-Means Clustering successfully delineated customer segments, highlighting a group with high incomes and a luxury preference as a strategic target for enhanced marketing efforts. Overall, the integrated use of these models has enabled a nuanced approach to customer segmentation, vital for optimizing marketing tactics and boosting sales in the supermarket.

9 of 12

Decision Tree

10 of 12

Classification performance table

11 of 12

Conclusion

Utilized "Adult Mortality Rate (2019-2021)" dataset for market clustering.
Employed three clusters based on the Elbow Method to avoid overfitting.
Clusters integrate health expenditures, GDP, and mortality for tailored products.
Segmentation enhances risk assessment and supports competitive pricing.
Insights improve operational efficiency and foster business growth.

The project utilized the "Adult Mortality Rate (2019-2021)" dataset to implement advanced clustering techniques, achieving a strategic segmentation of international health insurance markets based on key health, economic, and demographic indicators.

By employing three precisely determined clusters, as suggested by the Elbow Method analysis, the project successfully captured essential data structures without overfitting, allowing for customized insurance products.

Each cluster uniquely combines health expenditures, GDP per capita, and adult mortality rates, enabling the company to tailor its offerings to meet regional needs effectively.

This strategic segmentation has facilitated improved risk assessment and market differentiation, supporting the development of competitively priced insurance solutions that address the diverse requirements of global populations.

The insights gained from this project position the company to enhance operational efficiency, boost customer engagement, and drive growth, setting a foundation for strategic planning and informed decision-making in addressing health and economic challenges internationally.

12 of 12

Thank You!