Predicting Life Expectancy in Less Developed Countries

This research work was for my undergraduate thesis at Joseph Ayo Babalola University. It was based on selected Less Developed countries in Africa, Asia, and South America.

This repository contains the Jupyter notebook files, python files, and datasets used for this research work.

Aim/Objective ❔

This paper focuses on predicting life expectancy in selected less-developed countries. It also investigates the ways through which various factors affect life expectancy and the prediction strength of the same.

Methods

• Correlation Analysis
• Linear Regression machine learning model (for prediction). The model was trained and tested with the World Development Indicators dataset.
• 62 less developed countries were considered in all (selected based on their Human Development Index), and 12 features were used.
• Evaluation metrics and tests used include R-squared, Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE).

Results 🎯

✅ We found that Indicators thought to be great predictors of life expectancy were not. Strong predictors of life expectancy that were observed include literacy rate, infant mortality, sanitation, government expenditure on education, adolescent fertility rate, and percentage of urban population living in slums.

✅ The model had an accuracy of 79.55%

✅ The R-squared value was 0.8930.

✅ The Mean Absolute Error was 1.9190

✅ Mean Squared Error was 5.7623

✅ The Root Mean Squared Error was 2.4005

Libraries used

• NumPy
• Pandas
• Scikit-learn
• pycountry_convert
• statsmodel.

The initial dataset, WDIEXCEL.xlxs is the WHO's World Development Indicators dataset.

Notebook files 📁

• data_extracter.ipynb: This file contains code that work on the WDIEXCEL dataset and exports other Excel datasets based on it. These are the datasets that it produces, in order.
  1. WDI_EXCEL_TRIMMED.xlsx: This file trims the WDIEXCEL dataset to the following columns: Country Name, Indicator Name, Indicator Code, 2017, 2018, 2019
  2. WDI_EXCEL_TRIMMED_WITH_CONTINENT.xlsx: This file has the same contents as WDI_EXCEL_TRIMMED.xlsx, but with a new column, Continent.
  3. AFRICA_ASIA_SOUTHAME_CONTRIES.xlsx: This lists all the countries in Africa, Asia, and South America
  4. SELECTED_COUNTRIES_WITH_HDI: The code that produces this file appends to each country in AFRICA_ASIA_SOUTHAME_CONTRIES.xlsx, the corresponding Human Development Index (HDI) value.
• wdi_trim_to_selected_countries.ipynb: This notebook contains code that filters the WDI_EXCEL_TRIMMED_WITH_CONTINENT.xlsx file to only contain the countries in the SELECTED_COUNTRIES_WITH_HDI.xlsx file. It exports this data into WDI_EXCEL_TRIMMED_SELECTED_COUNTRIES.xlsx.
• filter_wdi_dataset_selected_variables.ipynb: This notebook produces a new Excel file, FINAL_FILTERED_WDI_DATASET.xlsx. This file contains the data in WDI_EXCEL_TRIMMED_SELECTED_COUNTRIES.xlsx filtered based on the initially selected indicators in the MAIN DATASET.xlsx, the Variables List worksheet.
• fill_dataset.ipynb: This notebook populates the FINAL_MAIN_TEST.xlsx, although I already made changes to the file by adding some new worksheets.
• linear_regression.ipynb: This notebook performs the Linear regression training and testing, and prints out the Model score, and evaluation metrics such as MAE, MSE, and R-squared. It uses the FINAL_DATASEST_CORRELATION.xlsx file which is a copy of Sheet 3 of the FINAL_MAIN_TEST.xlsx file. It uses the TrimmedIndicators work sheet. It also has other worksheets that shows the correlation analysis performed on the initially selected variables.
• regression_script.ipynb: This file performs the trains the linear regression model and prints out metrics using the statsmodel library.