Poverty-Analysis

This repository contains the analysis and modeling of poverty in Indonesia using statistical methods, particularly focusing on Geographically Weighted Regression (GWR). The study identifies influential factors and constructs predictive models to analyze poverty across Indonesia's 34 provinces.

Key Insights

Two datasets were analyzed, "Kemiskinan" and "Kemiskinan2", to determine which provides better insights into poverty modeling:

• Kemiskinan2 outperformed Kemiskinan in terms of accuracy, explanatory power, and interpretability.
• All subsequent analysis and modeling use the Kemiskinan2 dataset for its superior performance.

---

Repository Contents

1. Kemiskinan.xlsx and Kemiskinan2.xlsx:

   • Datasets containing poverty data across Indonesia's provinces.
   • Kemiskinan2 includes refined indicators for better model performance.

2. The Code.Rmd:

   • The R Markdown script for performing the statistical analysis and regression modeling.
   • Steps include data exploration, testing for multicollinearity, and building regression models (OLS, WLS, and GWR).

3. The Paper.docx:

   • Detailed documentation of the analysis, including methodology, results, and discussions.

---

Methodology

1. Models Used

• Ordinary Least Squares (OLS): Baseline linear regression model.
• Weighted Least Squares (WLS): Addresses heteroscedasticity issues in OLS.
• Geographically Weighted Regression (GWR): Incorporates spatial heterogeneity to improve model precision.

2. Evaluation Metrics

• R-Squared: Measures the proportion of variance explained by the model.
• AIC (Akaike Information Criterion): Evaluates model fit, penalizing for complexity.

---

Results: Why Kemiskinan2?

Model	Dataset	R-Squared	AIC	Remarks
GWR	Kemiskinan2	81.95%	530.18	Best fit; explains spatial variation effectively.
GWR	Kemiskinan	74.03%	537.70	Suboptimal; lower explanatory power.
OLS	Kemiskinan2	60.34%	559.59	Poor fit compared to GWR.
OLS	Kemiskinan	57.29%	556.11	Lowest fit; lacks spatial insights.

• Kemiskinan2 enables GWR to achieve the highest R-Squared and lowest AIC values, making it the most effective dataset for capturing spatial variability in poverty.

---

Steps to Reproduce

1. Install necessary R packages:

   install.packages(c("sp", "ggplot2", "GWmodel", "car"))

2. Open The Code.Rmd in RStudio or another compatible IDE.
3. Run the script to perform:
   • Data preprocessing
   • Model training and evaluation
   • Visualizations and result interpretation.

---

Conclusion

The GWR model using Kemiskinan2 is the most effective for analyzing poverty in Indonesia:

• Explains 81.95% of data variability (R-Squared).
• Provides actionable insights for targeted poverty reduction strategies.
• Highlights key variables such as education completion rate, adequate sanitation, and drinking water as critical to reducing poverty.

For further exploration, future research can incorporate advanced machine learning techniques or additional socioeconomic indicators to enhance the model.