---
title: "Readme"
author: "Gabriel Gonzalo Ojeda Cárcamo"
date: "16/12/2025"
output: pdf_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```

# Capstone CYO Project
## Unsupervised Clustering of International Tourists in Peru (FAMD + PAM)

This capstone project applies **unsupervised machine learning techniques** to segment international tourists visiting Peru using data from the **2024 Foreign Tourist Profile Survey (PROMPERÚ)**.
The analysis combines **Factor Analysis of Mixed Data (FAMD)** for dimensionality reduction with **Partitioning Around Medoids (PAM)** for clustering, enabling robust segmentation of heterogeneous survey data containing both categorical and numerical variables.

---

## Project Objectives

- Identify **latent tourist segments** based on sociodemographic, behavioral, and spending characteristics
- Apply **unsupervised learning methods** suitable for mixed-type data
- Validate clustering solutions using **multiple internal metrics**
- Assess cluster robustness using **bootstrap-based stability analysis**
- Generate **interpretable and actionable tourist profiles**

---

## Data Source

- **Survey:** Foreign Tourist Profile Survey - 2024
- **Institution:** PROMPERÚ
- **Coverage:** February, May, August, and November 2024
- **Sample size:** 5,268 international tourists
- **Location:** Jorge Chávez International Airport (Peru)
- **Population:** International tourists aged 15 years and older

The dataset includes detailed information on:
- Travel motivation
- Sociodemographic characteristics
- Planning behavior
- Occupation and income
- Total trip spending

---

## Methodological Framework

### 1. Data Preparation
- Selection of analytically relevant variables
- Conversion of SPSS-labelled variables to R factors
- Removal of variables with excessive missingness
- Log-transformation of total trip spending

### 2. Missing Value Imputation
- **missForest** (Random Forest-based imputation for mixed data)
- Strong performance: *NRMSE $\approx$ 0.19*, *PFC = 0.00*
- Preserves the multivariate structure of the data

### 3. Dimensionality Reduction
- **Factor Analysis of Mixed Data (FAMD)**
- Reduces dimensionality while retaining key sources of variance

### 4. Clustering
- **Partitioning Around Medoids (PAM)** applied to FAMD coordinates
- Distance-based clustering suitable for mixed-type data

### 5. Cluster Validation
- Silhouette Index
- Calinski-Harabasz Index
- Dunn Index
- Bootstrap stability analysis (Jaccard similarity)

A **five-cluster solution (k = 5)** was selected as the most balanced and interpretable configuration.

---

## Key Results

### Very young student leisure travelers
- Predominantly Centennials (15-24 years)
- Students, single, no children
- Leisure-oriented
- Moderate average spending

### Business-oriented working Millennials
- Business travel as main motive
- Short planning horizon
- Lowest average spending

### Mid-life families with higher spending
- Mainly Generation X
- Married with children
- High education levels
- High average trip expenditure

### Young adult leisure travelers
- Millennials and Centennials
- Leisure-focused
- Mostly private-sector workers
- Medium spending levels

### Senior high-income, high-spending travelers
- Baby Boomers (55+ years)
- Retired, highly educated
- Highest average spending

Cluster stability analysis produced **high Jaccard coefficients (0.84-0.98)**, confirming that the segments are robust and reproducible.

---

## Tools & Technologies

- **Language:** R
- **Core packages:**
  - `FactoMineR`, `factoextra`
  - `cluster`, `clusterCrit`, `fpc`
  - `missForest`
- **Data handling & visualization:**
  - `tidyverse`, `ggplot2`, `GGally`
- **Data import:**
  - `haven`, `sjlabelled`