"
+ ]
+ },
+ "metadata": {},
+ "output_type": "display_data"
+ }
+ ],
+ "source": [
+ "########################################################\n",
+ "## MATRICES DE CONFUSIÓN — AMBOS MODELOS ##\n",
+ "########################################################\n",
+ "\n",
+ "fig, axes = plt.subplots(1, 2, figsize=(13, 5))\n",
+ "\n",
+ "modelos_eval = [\n",
+ " (pred_val_knn, \"Mejor k-NN (Grid Search)\"),\n",
+ " (pred_val_log, \"Mejor Regresión Logística (Grid Search)\")\n",
+ "]\n",
+ "\n",
+ "for ax, (pred, titulo) in zip(axes, modelos_eval):\n",
+ " mc = confusion_matrix(y_val, pred)\n",
+ " sns.heatmap(\n",
+ " mc,\n",
+ " annot=True,\n",
+ " fmt=\"d\",\n",
+ " cmap=\"Blues\",\n",
+ " ax=ax,\n",
+ " annot_kws={\"size\": 14},\n",
+ " xticklabels=[\"No default (0)\", \"Default (1)\"],\n",
+ " yticklabels=[\"No default (0)\", \"Default (1)\"]\n",
+ " )\n",
+ " ax.set_title(titulo, fontsize=11, fontweight=\"bold\")\n",
+ " ax.set_xlabel(\"Predicción\", fontsize=10)\n",
+ " ax.set_ylabel(\"Valor real\", fontsize=10)\n",
+ "\n",
+ "plt.suptitle(\"Matrices de Confusión — Conjunto de Validación\",\n",
+ " fontsize=13, fontweight=\"bold\")\n",
+ "plt.tight_layout()\n",
+ "plt.show()"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "72c741c4",
+ "metadata": {},
+ "source": [
+ "Las matrices de confusión permiten comparar cómo se comportan los modelos al clasificar correctamente los casos de default y no default. En el caso de k-NN, el modelo clasifica correctamente 27 494 casos de no incumplimiento, pero solo identifica 378 de los 2 005 casos reales de default, lo que refleja una baja capacidad para detectar la clase minoritaria. Por otro lado, la Regresión Logística identifica 1 467 casos de default, una mejora sustancial frente a k-NN, aunque a costa de un mayor número de falsos positivos (clientes clasificados como default cuando no lo son). En problemas de riesgo crediticio, este comportamiento suele ser preferible, ya que es más costoso no detectar un cliente riesgoso (falso negativo) que generar una alerta adicional, lo que refuerza la elección de la Regresión Logística como modelo final."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### 6.3 Decisión del modelo final\n",
+ "\n",
+ "Con base en las métricas de validación cruzada (Sección 5) y la evaluación sobre el conjunto de validación (Sección 6), seleccionamos el modelo con mejor F1-Score en la clase 1 como **modelo final**.\n",
+ "\n",
+ "En el contexto de riesgo crediticio priorizamos el **Recall de la clase 1**: es preferible ser más cauteloso y señalar algunos falsos positivos que dejar pasar clientes que van a incumplir (falsos negativos)."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 28,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Comparación final de modelos — Conjunto de Validación:\n"
+ ]
+ },
+ {
+ "data": {
+ "text/html": [
+ "\n",
+ "\n",
+ "
\n",
+ " \n",
+ " \n",
+ " Modelo \n",
+ " Accuracy \n",
+ " Precisión (default) \n",
+ " Recall (default) \n",
+ " F1 (default) \n",
+ " \n",
+ " \n",
+ " \n",
+ " \n",
+ " 0 \n",
+ " k-NN (mejor Grid Search) \n",
+ " 0.9291 \n",
+ " 0.4300 \n",
+ " 0.1885 \n",
+ " 0.2621 \n",
+ " \n",
+ " \n",
+ " 1 \n",
+ " Regresión Logística (mejor Grid Search) \n",
+ " 0.8041 \n",
+ " 0.2155 \n",
+ " 0.7317 \n",
+ " 0.3330 \n",
+ " \n",
+ " \n",
+ "
\n",
+ "
\n",
+ "\n",
+ "
\n",
+ " \n",
+ " \n",
+ " RevolvingUtilizationOfUnsecuredLines \n",
+ " age \n",
+ " NumberOfTime30-59DaysPastDueNotWorse \n",
+ " NumberOfTimes90DaysLate \n",
+ " DebtRatio \n",
+ " MonthlyIncome \n",
+ " NumberOfOpenCreditLinesAndLoans \n",
+ " NumberRealEstateLoansOrLines \n",
+ " NumberOfDependents \n",
+ " flag_ingreso_nulo \n",
+ " flag_atrasos_98 \n",
+ " \n",
+ " \n",
+ " \n",
+ " \n",
+ " count \n",
+ " 101503.000000 \n",
+ " 101503.000000 \n",
+ " 101503.000000 \n",
+ " 101503.000000 \n",
+ " 101503.000000 \n",
+ " 101503.000000 \n",
+ " 101503.000000 \n",
+ " 101503.000000 \n",
+ " 101503.00000 \n",
+ " 101503.000000 \n",
+ " 101503.000000 \n",
+ " \n",
+ " \n",
+ " mean \n",
+ " 0.330565 \n",
+ " 52.404599 \n",
+ " 0.274524 \n",
+ " 0.117386 \n",
+ " 314.010870 \n",
+ " 6139.879521 \n",
+ " 8.453514 \n",
+ " 1.013074 \n",
+ " 0.74915 \n",
+ " 0.198053 \n",
+ " 0.002098 \n",
+ " \n",
+ " \n",
+ " std \n",
+ " 0.407885 \n",
+ " 14.777235 \n",
+ " 0.915546 \n",
+ " 0.758526 \n",
+ " 903.964523 \n",
+ " 3819.079148 \n",
+ " 5.144100 \n",
+ " 1.110253 \n",
+ " 1.12860 \n",
+ " 0.398534 \n",
+ " 0.045761 \n",
+ " \n",
+ " \n",
+ " min \n",
+ " 0.000000 \n",
+ " 21.000000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " 0.00000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " \n",
+ " \n",
+ " 25% \n",
+ " 0.030131 \n",
+ " 41.000000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " 0.173423 \n",
+ " 3900.000000 \n",
+ " 5.000000 \n",
+ " 0.000000 \n",
+ " 0.00000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " \n",
+ " \n",
+ " 50% \n",
+ " 0.152586 \n",
+ " 52.000000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " 0.364260 \n",
+ " 5400.000000 \n",
+ " 8.000000 \n",
+ " 1.000000 \n",
+ " 0.00000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " \n",
+ " \n",
+ " 75% \n",
+ " 0.564225 \n",
+ " 63.000000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " 0.851619 \n",
+ " 7374.000000 \n",
+ " 11.000000 \n",
+ " 2.000000 \n",
+ " 1.00000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " \n",
+ " \n",
+ " max \n",
+ " 5.000000 \n",
+ " 95.000000 \n",
+ " 13.000000 \n",
+ " 13.000000 \n",
+ " 4979.040000 \n",
+ " 23000.000000 \n",
+ " 85.000000 \n",
+ " 37.000000 \n",
+ " 43.00000 \n",
+ " 1.000000 \n",
+ " 1.000000 \n",
+ " \n",
+ " \n",
+ "
\n",
+ "
\n",
- "\n",
- "# Proyecto 1 \n",
- "## Clasificación de Riesgo Crediticio \n",
- "\n",
- "**Machine Learning** \n",
- "Universidad del Rosario\n",
- "\n",
- "Vanessa Ochoa, Nilson Amaya, Juan Zamora\n",
- "\n",
- "
\n"
- ]
- },
- {
- "cell_type": "markdown",
- "id": "02e6c89c",
- "metadata": {},
- "source": [
- "---"
- ]
- },
- {
- "cell_type": "markdown",
- "id": "8fc99973",
- "metadata": {},
- "source": [
- "## Tabla de contenido\n",
- "\n",
- "- [Introducción](#1-introducción)\n",
- "- [Importación del dataset](#2-importación-del-dataset)\n",
- "- [EDA](#3-análisis-exploratorio-de-datos-eda)\n",
- "- [Modelos](#4-modelos)\n",
- "- [Resultados](#5-resultados)\n",
- "- [Conclusiones](#6-conclusiones)"
- ]
- },
- {
- "cell_type": "markdown",
- "id": "03f0d790",
- "metadata": {},
- "source": [
- "\n",
- "RangeIndex: 150000 entries, 0 to 149999\n",
- "Data columns (total 12 columns):\n",
- " # Column Non-Null Count Dtype \n",
- "--- ------ -------------- ----- \n",
- " 0 Unnamed: 0 150000 non-null int64 \n",
- " 1 SeriousDlqin2yrs 150000 non-null int64 \n",
- " 2 RevolvingUtilizationOfUnsecuredLines 150000 non-null float64\n",
- " 3 age 150000 non-null int64 \n",
- " 4 NumberOfTime30-59DaysPastDueNotWorse 150000 non-null int64 \n",
- " 5 DebtRatio 150000 non-null float64\n",
- " 6 MonthlyIncome 120269 non-null float64\n",
- " 7 NumberOfOpenCreditLinesAndLoans 150000 non-null int64 \n",
- " 8 NumberOfTimes90DaysLate 150000 non-null int64 \n",
- " 9 NumberRealEstateLoansOrLines 150000 non-null int64 \n",
- " 10 NumberOfTime60-89DaysPastDueNotWorse 150000 non-null int64 \n",
- " 11 NumberOfDependents 146076 non-null float64\n",
- "dtypes: float64(4), int64(8)\n",
- "memory usage: 13.7 MB\n"
- ]
+ "cells": [
+ {
+ "cell_type": "markdown",
+ "id": "b1ca7847",
+ "metadata": {},
+ "source": [
+ "\n",
+ "\n",
+ "# Proyecto 1 \n",
+ "## Clasificación de Riesgo Crediticio \n",
+ "\n",
+ "**Machine Learning** \n",
+ "Universidad del Rosario\n",
+ "\n",
+ "Vanessa Ochoa, Nilson Amaya, Juan Zamora\n",
+ "\n",
+ "
\n"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "02e6c89c",
+ "metadata": {},
+ "source": [
+ "---"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "8fc99973",
+ "metadata": {},
+ "source": [
+ "## Tabla de contenido\n",
+ "\n",
+ "- [Introducción](#1-introducción)\n",
+ "- [Importación del dataset](#2-importación-del-dataset)\n",
+ "- [EDA](#3-análisis-exploratorio-de-datos-eda)\n",
+ "- [Modelos](#4-modelos)\n",
+ "- [Optimización de hiperparámetros y validación](#5-resultados)\n",
+ "- [Métricas de evaluación, ROC y decisión de umbral](#6-conclusiones)\n",
+ "- [Predicción en el conjunto de test](#7-test)\n",
+ "- [Diagrama del pipeline](#8-diagrama)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "03f0d790",
+ "metadata": {},
+ "source": [
+ "\n",
+ "RangeIndex: 150000 entries, 0 to 149999\n",
+ "Data columns (total 11 columns):\n",
+ " # Column Non-Null Count Dtype \n",
+ "--- ------ -------------- ----- \n",
+ " 0 SeriousDlqin2yrs 150000 non-null int64 \n",
+ " 1 RevolvingUtilizationOfUnsecuredLines 150000 non-null float64\n",
+ " 2 age 150000 non-null int64 \n",
+ " 3 NumberOfTime30-59DaysPastDueNotWorse 150000 non-null int64 \n",
+ " 4 DebtRatio 150000 non-null float64\n",
+ " 5 MonthlyIncome 120269 non-null float64\n",
+ " 6 NumberOfOpenCreditLinesAndLoans 150000 non-null int64 \n",
+ " 7 NumberOfTimes90DaysLate 150000 non-null int64 \n",
+ " 8 NumberRealEstateLoansOrLines 150000 non-null int64 \n",
+ " 9 NumberOfTime60-89DaysPastDueNotWorse 150000 non-null int64 \n",
+ " 10 NumberOfDependents 146076 non-null float64\n",
+ "dtypes: float64(4), int64(7)\n",
+ "memory usage: 12.6 MB\n"
+ ]
+ }
+ ],
+ "source": [
+ "# Cargamos el dataset\n",
+ "\n",
+ "train = pd.read_csv(\"../input/cs-training.csv\")\n",
+ "test = pd.read_csv(\"../input/cs-test.csv\")\n",
+ "train = train.drop(columns=['Unnamed: 0'])\n",
+ "test = test.drop(columns=['Unnamed: 0'])\n",
+ "\n",
+ "# ver primeras filas\n",
+ "train.info()"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "099d50cc",
+ "metadata": {},
+ "source": [
+ "### 2.1 Descripción de las variables\n",
+ "\n",
+ "- **SeriousDlqin2yrs:** Indica si la persona ha tenido una morosidad grave, es decir, un atraso de 90 días o más en el pago de sus obligaciones durante los últimos dos años. Es la variable objetivo del modelo.\n",
+ "\n",
+ "- **RevolvingUtilizationOfUnsecuredLines:** Proporción del saldo utilizado en tarjetas de crédito y líneas de crédito personales respecto al límite total disponible.\n",
+ "\n",
+ "- **age:** Edad del solicitante del crédito expresada en años.\n",
+ "\n",
+ "- **NumberOfTime30-59DaysPastDueNotWorse:** Número de veces que el prestatario ha tenido retrasos en sus pagos entre 30 y 59 días durante los últimos dos años.\n",
+ "\n",
+ "- **DebtRatio:** Relación entre las obligaciones financieras mensuales (deudas, manutención, gastos) y el ingreso bruto mensual del prestatario.\n",
+ "\n",
+ "- **MonthlyIncome:** Ingreso mensual reportado por el prestatario.\n",
+ "\n",
+ "- **NumberOfOpenCreditLinesAndLoans:** Número total de préstamos y líneas de crédito abiertas, como tarjetas de crédito, préstamos de auto o hipotecas.\n",
+ "\n",
+ "- **NumberOfTimes90DaysLate:** Número de veces que el prestatario ha tenido retrasos en pagos de 90 días o más.\n",
+ "\n",
+ "- **NumberRealEstateLoansOrLines:** Número de préstamos relacionados con bienes raíces, incluyendo hipotecas y líneas de crédito sobre vivienda.\n",
+ "\n",
+ "- **NumberOfTime60-89DaysPastDueNotWorse:** Número de veces que el prestatario ha tenido retrasos en pagos entre 60 y 89 días en los últimos dos años.\n",
+ "\n",
+ "- **NumberOfDependents:** Número de personas dependientes económicamente del prestatario, como hijos o cónyuge."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 3,
+ "id": "7b59092d",
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/html": [
+ "\n",
+ "\n",
+ "
\n",
+ " \n",
+ " \n",
+ " SeriousDlqin2yrs \n",
+ " RevolvingUtilizationOfUnsecuredLines \n",
+ " age \n",
+ " NumberOfTime30-59DaysPastDueNotWorse \n",
+ " DebtRatio \n",
+ " MonthlyIncome \n",
+ " NumberOfOpenCreditLinesAndLoans \n",
+ " NumberOfTimes90DaysLate \n",
+ " NumberRealEstateLoansOrLines \n",
+ " NumberOfTime60-89DaysPastDueNotWorse \n",
+ " NumberOfDependents \n",
+ " \n",
+ " \n",
+ " \n",
+ " \n",
+ " count \n",
+ " 150000.000000 \n",
+ " 150000.000000 \n",
+ " 150000.000000 \n",
+ " 150000.000000 \n",
+ " 150000.000000 \n",
+ " 1.202690e+05 \n",
+ " 150000.000000 \n",
+ " 150000.000000 \n",
+ " 150000.000000 \n",
+ " 150000.000000 \n",
+ " 146076.000000 \n",
+ " \n",
+ " \n",
+ " mean \n",
+ " 0.066840 \n",
+ " 6.048438 \n",
+ " 52.295207 \n",
+ " 0.421033 \n",
+ " 353.005076 \n",
+ " 6.670221e+03 \n",
+ " 8.452760 \n",
+ " 0.265973 \n",
+ " 1.018240 \n",
+ " 0.240387 \n",
+ " 0.757222 \n",
+ " \n",
+ " \n",
+ " std \n",
+ " 0.249746 \n",
+ " 249.755371 \n",
+ " 14.771866 \n",
+ " 4.192781 \n",
+ " 2037.818523 \n",
+ " 1.438467e+04 \n",
+ " 5.145951 \n",
+ " 4.169304 \n",
+ " 1.129771 \n",
+ " 4.155179 \n",
+ " 1.115086 \n",
+ " \n",
+ " \n",
+ " min \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " 0.000000e+00 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " \n",
+ " \n",
+ " 25% \n",
+ " 0.000000 \n",
+ " 0.029867 \n",
+ " 41.000000 \n",
+ " 0.000000 \n",
+ " 0.175074 \n",
+ " 3.400000e+03 \n",
+ " 5.000000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " \n",
+ " \n",
+ " 50% \n",
+ " 0.000000 \n",
+ " 0.154181 \n",
+ " 52.000000 \n",
+ " 0.000000 \n",
+ " 0.366508 \n",
+ " 5.400000e+03 \n",
+ " 8.000000 \n",
+ " 0.000000 \n",
+ " 1.000000 \n",
+ " 0.000000 \n",
+ " 0.000000 \n",
+ " \n",
+ " \n",
+ " 75% \n",
+ " 0.000000 \n",
+ " 0.559046 \n",
+ " 63.000000 \n",
+ " 0.000000 \n",
+ " 0.868254 \n",
+ " 8.249000e+03 \n",
+ " 11.000000 \n",
+ " 0.000000 \n",
+ " 2.000000 \n",
+ " 0.000000 \n",
+ " 1.000000 \n",
+ " \n",
+ " \n",
+ " max \n",
+ " 1.000000 \n",
+ " 50708.000000 \n",
+ " 109.000000 \n",
+ " 98.000000 \n",
+ " 329664.000000 \n",
+ " 3.008750e+06 \n",
+ " 58.000000 \n",
+ " 98.000000 \n",
+ " 54.000000 \n",
+ " 98.000000 \n",
+ " 20.000000 \n",
+ " \n",
+ " \n",
+ "
\n",
+ "
"
+ ]
+ },
+ "metadata": {},
+ "output_type": "display_data"
+ }
+ ],
+ "source": [
+ "fig, axes = plt.subplots(3, 3, figsize=(15, 12))\n",
+ "axes = axes.flatten()\n",
+ "\n",
+ "columnas = [\n",
+ " 'RevolvingUtilizationOfUnsecuredLines', 'age',\n",
+ " 'NumberOfTime30-59DaysPastDueNotWorse', 'DebtRatio',\n",
+ " 'MonthlyIncome', 'NumberOfOpenCreditLinesAndLoans',\n",
+ " 'NumberOfTimes90DaysLate', 'NumberOfDependents',\n",
+ " 'NumberOfTime60-89DaysPastDueNotWorse'\n",
+ "]\n",
+ "\n",
+ "for i, col in enumerate(columnas):\n",
+ " axes[i].boxplot(train[col], patch_artist=True,\n",
+ " boxprops=dict(facecolor='lightblue', color='navy'),\n",
+ " medianprops=dict(color='red'))\n",
+ " \n",
+ " axes[i].set_title(col, fontsize=10, fontweight='bold')\n",
+ " axes[i].set_ylabel('Valor')\n",
+ " axes[i].grid(alpha=0.3)\n",
+ "\n",
+ "fig.suptitle('Detección de Outliers por Variable',\n",
+ " fontsize=16, fontweight='bold')\n",
+ "\n",
+ "plt.tight_layout()\n",
+ "plt.show()"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "4cd7a8ed",
+ "metadata": {},
+ "source": [
+ "A partir de los boxplots se observa que varias variables presentan valores atípicos, especialmente en RevolvingUtilizationOfUnsecuredLines, DebtRatio y MonthlyIncome, donde hay una gran dispersión y algunos valores extremadamente altos. En las variables relacionadas con retrasos en pagos, la mayoría de los valores se concentran cerca de cero, pero aparecen algunos casos con muchos incumplimientos."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 12,
+ "id": "ce567172",
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Outliers tratados en variables continuas\n",
+ "\n",
+ "Estadísticas después del tratamiento:\n"
+ ]
+ },
+ {
+ "data": {
+ "text/html": [
+ "\n",
+ "\n",
+ "
\n",
+ " \n",
+ " \n",
+ " SeriousDlqin2yrs \n",
+ " RevolvingUtilizationOfUnsecuredLines \n",
+ " age \n",
+ " NumberOfTime30-59DaysPastDueNotWorse \n",
+ " DebtRatio \n",
+ " MonthlyIncome \n",
+ " NumberOfOpenCreditLinesAndLoans \n",
+ " NumberOfTimes90DaysLate \n",
+ " NumberRealEstateLoansOrLines \n",
+ " NumberOfTime60-89DaysPastDueNotWorse \n",
+ " NumberOfDependents \n",
+ " \n",
+ " \n",
+ " \n",
+ " \n",
+ " count \n",
+ " 150000.00 \n",
+ " 150000.00 \n",
+ " 150000.00 \n",
+ " 150000.00 \n",
+ " 150000.00 \n",
+ " 150000.00 \n",
+ " 150000.00 \n",
+ " 150000.00 \n",
+ " 150000.00 \n",
+ " 150000.00 \n",
+ " 150000.00 \n",
+ " \n",
+ " \n",
+ " mean \n",
+ " 0.07 \n",
+ " 0.32 \n",
+ " 52.30 \n",
+ " 0.42 \n",
+ " 0.47 \n",
+ " 6142.45 \n",
+ " 8.45 \n",
+ " 0.27 \n",
+ " 1.02 \n",
+ " 0.24 \n",
+ " 0.74 \n",
+ " \n",
+ " \n",
+ " std \n",
+ " 0.25 \n",
+ " 0.35 \n",
+ " 14.77 \n",
+ " 4.19 \n",
+ " 0.36 \n",
+ " 3835.68 \n",
+ " 5.15 \n",
+ " 4.17 \n",
+ " 1.13 \n",
+ " 4.16 \n",
+ " 1.11 \n",
+ " \n",
+ " \n",
+ " min \n",
+ " 0.00 \n",
+ " 0.00 \n",
+ " 18.00 \n",
+ " 0.00 \n",
+ " 0.00 \n",
+ " 0.00 \n",
+ " 0.00 \n",
+ " 0.00 \n",
+ " 0.00 \n",
+ " 0.00 \n",
+ " 0.00 \n",
+ " \n",
+ " \n",
+ " 25% \n",
+ " 0.00 \n",
+ " 0.03 \n",
+ " 41.00 \n",
+ " 0.00 \n",
+ " 0.18 \n",
+ " 3903.00 \n",
+ " 5.00 \n",
+ " 0.00 \n",
+ " 0.00 \n",
+ " 0.00 \n",
+ " 0.00 \n",
+ " \n",
+ " \n",
+ " 50% \n",
+ " 0.00 \n",
+ " 0.15 \n",
+ " 52.00 \n",
+ " 0.00 \n",
+ " 0.37 \n",
+ " 5400.00 \n",
+ " 8.00 \n",
+ " 0.00 \n",
+ " 1.00 \n",
+ " 0.00 \n",
+ " 0.00 \n",
+ " \n",
+ " \n",
+ " 75% \n",
+ " 0.00 \n",
+ " 0.56 \n",
+ " 63.00 \n",
+ " 0.00 \n",
+ " 0.87 \n",
+ " 7400.00 \n",
+ " 11.00 \n",
+ " 0.00 \n",
+ " 2.00 \n",
+ " 0.00 \n",
+ " 1.00 \n",
+ " \n",
+ " \n",
+ " max \n",
+ " 1.00 \n",
+ " 1.00 \n",
+ " 100.00 \n",
+ " 98.00 \n",
+ " 1.00 \n",
+ " 23000.00 \n",
+ " 58.00 \n",
+ " 98.00 \n",
+ " 54.00 \n",
+ " 98.00 \n",
+ " 20.00 \n",
+ " \n",
+ " \n",
+ "
\n",
+ "
"
+ ]
+ },
+ "metadata": {},
+ "output_type": "display_data"
+ },
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "\n",
+ "Correlación con SeriousDlqin2yrs:\n",
+ "SeriousDlqin2yrs 1.000000\n",
+ "NumberOfTime30-59DaysPastDueNotWorse 0.125587\n",
+ "NumberOfTimes90DaysLate 0.117175\n",
+ "NumberOfTime60-89DaysPastDueNotWorse 0.102261\n",
+ "NumberOfDependents 0.046869\n",
+ "RevolvingUtilizationOfUnsecuredLines -0.001802\n",
+ "NumberRealEstateLoansOrLines -0.007038\n",
+ "DebtRatio -0.007602\n",
+ "MonthlyIncome -0.017151\n",
+ "NumberOfOpenCreditLinesAndLoans -0.029669\n",
+ "age -0.115386\n",
+ "Name: SeriousDlqin2yrs, dtype: float64\n"
+ ]
+ }
+ ],
+ "source": [
+ "# Mapa de calor de correlaciones\n",
+ "plt.figure(figsize=(12, 8))\n",
+ "correlaciones = train.corr()\n",
+ "\n",
+ "sns.heatmap(correlaciones,\n",
+ " annot=True, # muestra los números\n",
+ " fmt='.2f', # 2 decimales\n",
+ " cmap='coolwarm', # azul=negativo, rojo=positivo\n",
+ " center=0,\n",
+ " square=True)\n",
+ "\n",
+ "plt.title('Mapa de Correlaciones', fontsize=14)\n",
+ "plt.tight_layout()\n",
+ "plt.show()\n",
+ "\n",
+ "# Correlaciones con la variable objetivo\n",
+ "print(\"\\nCorrelación con SeriousDlqin2yrs:\")\n",
+ "print(correlaciones['SeriousDlqin2yrs'].sort_values(ascending=False))"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "f64323be",
+ "metadata": {},
+ "source": [
+ "Para el análisis se utilizaron únicamente las variables con mayor relación con el incumplimiento (SeriousDlqin2yrs). Dado que las variables de retrasos en pagos presentan correlaciones muy altas entre sí, se decidió no considerar NumberOfTime60-89DaysPastDueNotWorse para evitar redundancia y posibles problemas de multicolinealidad."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 7,
+ "id": "ad52d012",
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Variables seleccionadas para modelado:\n",
+ "['RevolvingUtilizationOfUnsecuredLines', 'age', 'NumberOfTime30-59DaysPastDueNotWorse', 'NumberOfTimes90DaysLate', 'DebtRatio', 'MonthlyIncome', 'NumberOfOpenCreditLinesAndLoans', 'NumberRealEstateLoansOrLines', 'NumberOfDependents']\n",
+ "\n",
+ "Dimensiones de X e y: (150000, 9) (150000,)\n"
+ ]
+ }
+ ],
+ "source": [
+ "# Selección justificada de características\n",
+ "variables = [\n",
+ " 'RevolvingUtilizationOfUnsecuredLines',\n",
+ " 'age',\n",
+ " 'NumberOfTime30-59DaysPastDueNotWorse',\n",
+ " 'NumberOfTimes90DaysLate',\n",
+ " 'DebtRatio',\n",
+ " 'MonthlyIncome',\n",
+ " 'NumberOfOpenCreditLinesAndLoans',\n",
+ " 'NumberRealEstateLoansOrLines',\n",
+ " 'NumberOfDependents'\n",
+ "]\n",
+ "\n",
+ "X = train[variables]\n",
+ "y = train['SeriousDlqin2yrs']\n",
+ "\n",
+ "print('Variables seleccionadas para modelado:')\n",
+ "print(variables)\n",
+ "print('\\nDimensiones de X e y:', X.shape, y.shape)\n"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "a7b34b7c",
+ "metadata": {},
+ "source": [
+ "