Modulo 1

# install.packages

library(ggplot2)
library(lattice)
library(caret)
library(DataExplorer)
library(dplyr)

## 
## Attaching package: 'dplyr'

## The following objects are masked from 'package:stats':
## 
##     filter, lag

## The following objects are masked from 'package:base':
## 
##     intersect, setdiff, setequal, union

Crear la base de datos

  df <- read.csv("/Users/nataliamartinez/Desktop/M1_data.csv")
# Vista rápida
dim(df)

## [1] 133  22

head(df)

##   trust_apple interest_computers age_computer user_pcmac appleproducts_count
## 1          No                  4            8         PC                   0
## 2         Yes                  2            4         PC                   1
## 3         Yes                  5            6         PC                   0
## 4         Yes                  2            6      Apple                   4
## 5         Yes                  4            4      Apple                   7
## 6         Yes                  3            1      Apple                   2
##   familiarity_m1 f_batterylife f_price f_size f_multitasking f_noise
## 1             No             5       4      3              4       4
## 2             No             5       5      5              3       4
## 3             No             3       4      2              4       1
## 4             No             4       3      3              4       4
## 5            Yes             5       3      3              4       4
## 6             No             5       5      4              4       5
##   f_performance f_neural f_synergy f_performanceloss m1_consideration
## 1             2        2         1                 1                1
## 2             5        2         2                 4                2
## 3             4        2         2                 2                4
## 4             4        4         4                 3                2
## 5             5        3         4                 4                4
## 6             5        5         4                 2                2
##   m1_purchase gender age_group income_group   status          domain
## 1         Yes   Male         2            2  Student         Science
## 2          No   Male         2            3 Employed         Finance
## 3         Yes   Male         2            2  Student IT & Technology
## 4          No Female         2            2  Student  Arts & Culture
## 5         Yes   Male         5            7 Employed     Hospitality
## 6          No Female         2            2  Student        Politics

Entender la base de datos

# 1) Convertir strings a factor 
df <- df %>% mutate(across(where(is.character), as.factor))

# 2) Target como factor
df$m1_purchase <- as.factor(df$m1_purchase)

# 3) Quitar filas incompletas
df <- na.omit(df)

# 4) Ver balance de clases
table(df$m1_purchase)

## 
##  No Yes 
##  45  88

prop.table(table(df$m1_purchase))

## 
##        No       Yes 
## 0.3383459 0.6616541

Partir la base de datos

set.seed(123)
renglones_entrenamiento <- createDataPartition(df$m1_purchase, p = 0.8, list = FALSE)

entrenamiento <- df[renglones_entrenamiento, ]
prueba <- df[-renglones_entrenamiento, ]

dim(entrenamiento)

## [1] 107  22

dim(prueba)

## [1] 26 22

Distintos tipos de Métodos para Modelar

ctrl <- trainControl(method = "cv", number = 10)

Modelo 1. SVM Lineal

set.seed(123)
modelo1 <- train(
  m1_purchase ~ .,
  data = entrenamiento,
  method = "svmLinear",
  preProcess = c("center", "scale"),
  trControl = ctrl,
  tuneGrid = data.frame(C = 1)
)

resultado_entrenamiento1 <- predict(modelo1, entrenamiento)
resultado_prueba1 <- predict(modelo1, prueba)

mcre1 <- confusionMatrix(resultado_entrenamiento1, entrenamiento$m1_purchase)
mcrp1 <- confusionMatrix(resultado_prueba1, prueba$m1_purchase)

mcre1

## Confusion Matrix and Statistics
## 
##           Reference
## Prediction No Yes
##        No  30   4
##        Yes  6  67
##                                           
##                Accuracy : 0.9065          
##                  95% CI : (0.8348, 0.9543)
##     No Information Rate : 0.6636          
##     P-Value [Acc > NIR] : 4.281e-09       
##                                           
##                   Kappa : 0.7878          
##                                           
##  Mcnemar's Test P-Value : 0.7518          
##                                           
##             Sensitivity : 0.8333          
##             Specificity : 0.9437          
##          Pos Pred Value : 0.8824          
##          Neg Pred Value : 0.9178          
##              Prevalence : 0.3364          
##          Detection Rate : 0.2804          
##    Detection Prevalence : 0.3178          
##       Balanced Accuracy : 0.8885          
##                                           
##        'Positive' Class : No              
## 

mcrp1

## Confusion Matrix and Statistics
## 
##           Reference
## Prediction No Yes
##        No   3   6
##        Yes  6  11
##                                           
##                Accuracy : 0.5385          
##                  95% CI : (0.3337, 0.7341)
##     No Information Rate : 0.6538          
##     P-Value [Acc > NIR] : 0.9231          
##                                           
##                   Kappa : -0.0196         
##                                           
##  Mcnemar's Test P-Value : 1.0000          
##                                           
##             Sensitivity : 0.3333          
##             Specificity : 0.6471          
##          Pos Pred Value : 0.3333          
##          Neg Pred Value : 0.6471          
##              Prevalence : 0.3462          
##          Detection Rate : 0.1154          
##    Detection Prevalence : 0.3462          
##       Balanced Accuracy : 0.4902          
##                                           
##        'Positive' Class : No              
## 

Modelo 2. SVM Radial

set.seed(123)
modelo2 <- train(
  m1_purchase ~ .,
  data = entrenamiento,
  method = "svmRadial",
  preProcess = c("center", "scale"),
  trControl = ctrl,
  tuneGrid = data.frame(sigma = 0.1, C = 1)  # puedes ajustar sigma si quieres
)

resultado_entrenamiento2 <- predict(modelo2, entrenamiento)
resultado_prueba2 <- predict(modelo2, prueba)

mcre2 <- confusionMatrix(resultado_entrenamiento2, entrenamiento$m1_purchase)
mcrp2 <- confusionMatrix(resultado_prueba2, prueba$m1_purchase)

mcre2

## Confusion Matrix and Statistics
## 
##           Reference
## Prediction No Yes
##        No  34   0
##        Yes  2  71
##                                           
##                Accuracy : 0.9813          
##                  95% CI : (0.9341, 0.9977)
##     No Information Rate : 0.6636          
##     P-Value [Acc > NIR] : <2e-16          
##                                           
##                   Kappa : 0.9576          
##                                           
##  Mcnemar's Test P-Value : 0.4795          
##                                           
##             Sensitivity : 0.9444          
##             Specificity : 1.0000          
##          Pos Pred Value : 1.0000          
##          Neg Pred Value : 0.9726          
##              Prevalence : 0.3364          
##          Detection Rate : 0.3178          
##    Detection Prevalence : 0.3178          
##       Balanced Accuracy : 0.9722          
##                                           
##        'Positive' Class : No              
## 

mcrp2

## Confusion Matrix and Statistics
## 
##           Reference
## Prediction No Yes
##        No   0   0
##        Yes  9  17
##                                           
##                Accuracy : 0.6538          
##                  95% CI : (0.4433, 0.8279)
##     No Information Rate : 0.6538          
##     P-Value [Acc > NIR] : 0.589398        
##                                           
##                   Kappa : 0               
##                                           
##  Mcnemar's Test P-Value : 0.007661        
##                                           
##             Sensitivity : 0.0000          
##             Specificity : 1.0000          
##          Pos Pred Value :    NaN          
##          Neg Pred Value : 0.6538          
##              Prevalence : 0.3462          
##          Detection Rate : 0.0000          
##    Detection Prevalence : 0.0000          
##       Balanced Accuracy : 0.5000          
##                                           
##        'Positive' Class : No              
## 

Modelo 3. SVM Polinómico

set.seed(123)
modelo3 <- train(
  m1_purchase ~ .,
  data = entrenamiento,
  method = "svmPoly",
  preProcess = c("center", "scale"),
  trControl = ctrl,
  tuneGrid = data.frame(degree = 2, scale = 1, C = 1)
)

resultado_entrenamiento3 <- predict(modelo3, entrenamiento)
resultado_prueba3 <- predict(modelo3, prueba)

mcre3 <- confusionMatrix(resultado_entrenamiento3, entrenamiento$m1_purchase)
mcrp3 <- confusionMatrix(resultado_prueba3, prueba$m1_purchase)

mcre3

## Confusion Matrix and Statistics
## 
##           Reference
## Prediction No Yes
##        No  34   0
##        Yes  2  71
##                                           
##                Accuracy : 0.9813          
##                  95% CI : (0.9341, 0.9977)
##     No Information Rate : 0.6636          
##     P-Value [Acc > NIR] : <2e-16          
##                                           
##                   Kappa : 0.9576          
##                                           
##  Mcnemar's Test P-Value : 0.4795          
##                                           
##             Sensitivity : 0.9444          
##             Specificity : 1.0000          
##          Pos Pred Value : 1.0000          
##          Neg Pred Value : 0.9726          
##              Prevalence : 0.3364          
##          Detection Rate : 0.3178          
##    Detection Prevalence : 0.3178          
##       Balanced Accuracy : 0.9722          
##                                           
##        'Positive' Class : No              
## 

mcrp3

## Confusion Matrix and Statistics
## 
##           Reference
## Prediction No Yes
##        No   1   4
##        Yes  8  13
##                                           
##                Accuracy : 0.5385          
##                  95% CI : (0.3337, 0.7341)
##     No Information Rate : 0.6538          
##     P-Value [Acc > NIR] : 0.9231          
##                                           
##                   Kappa : -0.1387         
##                                           
##  Mcnemar's Test P-Value : 0.3865          
##                                           
##             Sensitivity : 0.11111         
##             Specificity : 0.76471         
##          Pos Pred Value : 0.20000         
##          Neg Pred Value : 0.61905         
##              Prevalence : 0.34615         
##          Detection Rate : 0.03846         
##    Detection Prevalence : 0.19231         
##       Balanced Accuracy : 0.43791         
##                                           
##        'Positive' Class : No              
## 

Modelo 4. Árbol de Decisión

set.seed(123)
modelo4 <- train(
  m1_purchase ~ .,
  data = entrenamiento,
  method = "rpart",
  trControl = ctrl,
  tuneLength = 10
)

resultado_entrenamiento4 <- predict(modelo4, entrenamiento)
resultado_prueba4 <- predict(modelo4, prueba)

mcre4 <- confusionMatrix(resultado_entrenamiento4, entrenamiento$m1_purchase)
mcrp4 <- confusionMatrix(resultado_prueba4, prueba$m1_purchase)

mcre4

## Confusion Matrix and Statistics
## 
##           Reference
## Prediction No Yes
##        No  17   2
##        Yes 19  69
##                                           
##                Accuracy : 0.8037          
##                  95% CI : (0.7158, 0.8742)
##     No Information Rate : 0.6636          
##     P-Value [Acc > NIR] : 0.0010139       
##                                           
##                   Kappa : 0.5025          
##                                           
##  Mcnemar's Test P-Value : 0.0004803       
##                                           
##             Sensitivity : 0.4722          
##             Specificity : 0.9718          
##          Pos Pred Value : 0.8947          
##          Neg Pred Value : 0.7841          
##              Prevalence : 0.3364          
##          Detection Rate : 0.1589          
##    Detection Prevalence : 0.1776          
##       Balanced Accuracy : 0.7220          
##                                           
##        'Positive' Class : No              
## 

mcrp4

## Confusion Matrix and Statistics
## 
##           Reference
## Prediction No Yes
##        No   4   6
##        Yes  5  11
##                                           
##                Accuracy : 0.5769          
##                  95% CI : (0.3692, 0.7665)
##     No Information Rate : 0.6538          
##     P-Value [Acc > NIR] : 0.8485          
##                                           
##                   Kappa : 0.0892          
##                                           
##  Mcnemar's Test P-Value : 1.0000          
##                                           
##             Sensitivity : 0.4444          
##             Specificity : 0.6471          
##          Pos Pred Value : 0.4000          
##          Neg Pred Value : 0.6875          
##              Prevalence : 0.3462          
##          Detection Rate : 0.1538          
##    Detection Prevalence : 0.3846          
##       Balanced Accuracy : 0.5458          
##                                           
##        'Positive' Class : No              
## 

Modelo 5. Redes Neuronales

set.seed(123)
modelo5 <- train(
  m1_purchase ~ .,
  data = entrenamiento,
  method = "nnet",
  preProcess = c("center", "scale"),
  trControl = ctrl,
  tuneLength = 10,
  trace = FALSE
)

resultado_entrenamiento5 <- predict(modelo5, entrenamiento)
resultado_prueba5 <- predict(modelo5, prueba)

mcre5 <- confusionMatrix(resultado_entrenamiento5, entrenamiento$m1_purchase)
mcrp5 <- confusionMatrix(resultado_prueba5, prueba$m1_purchase)

mcre5

## Confusion Matrix and Statistics
## 
##           Reference
## Prediction No Yes
##        No  34   0
##        Yes  2  71
##                                           
##                Accuracy : 0.9813          
##                  95% CI : (0.9341, 0.9977)
##     No Information Rate : 0.6636          
##     P-Value [Acc > NIR] : <2e-16          
##                                           
##                   Kappa : 0.9576          
##                                           
##  Mcnemar's Test P-Value : 0.4795          
##                                           
##             Sensitivity : 0.9444          
##             Specificity : 1.0000          
##          Pos Pred Value : 1.0000          
##          Neg Pred Value : 0.9726          
##              Prevalence : 0.3364          
##          Detection Rate : 0.3178          
##    Detection Prevalence : 0.3178          
##       Balanced Accuracy : 0.9722          
##                                           
##        'Positive' Class : No              
## 

mcrp5

## Confusion Matrix and Statistics
## 
##           Reference
## Prediction No Yes
##        No   3   9
##        Yes  6   8
##                                           
##                Accuracy : 0.4231          
##                  95% CI : (0.2335, 0.6308)
##     No Information Rate : 0.6538          
##     P-Value [Acc > NIR] : 0.9954          
##                                           
##                   Kappa : -0.1818         
##                                           
##  Mcnemar's Test P-Value : 0.6056          
##                                           
##             Sensitivity : 0.3333          
##             Specificity : 0.4706          
##          Pos Pred Value : 0.2500          
##          Neg Pred Value : 0.5714          
##              Prevalence : 0.3462          
##          Detection Rate : 0.1154          
##    Detection Prevalence : 0.4615          
##       Balanced Accuracy : 0.4020          
##                                           
##        'Positive' Class : No              
## 

Modelo 6. Bosques Aleatorios

set.seed(123)
modelo6 <- train(
  m1_purchase ~ .,
  data = entrenamiento,
  method = "rf",
  trControl = ctrl,
  tuneGrid = expand.grid(mtry = c(2, 4, 6))
)

resultado_entrenamiento6 <- predict(modelo6, entrenamiento)
resultado_prueba6 <- predict(modelo6, prueba)

mcre6 <- confusionMatrix(resultado_entrenamiento6, entrenamiento$m1_purchase)
mcrp6 <- confusionMatrix(resultado_prueba6, prueba$m1_purchase)

mcre6

## Confusion Matrix and Statistics
## 
##           Reference
## Prediction No Yes
##        No  34   0
##        Yes  2  71
##                                           
##                Accuracy : 0.9813          
##                  95% CI : (0.9341, 0.9977)
##     No Information Rate : 0.6636          
##     P-Value [Acc > NIR] : <2e-16          
##                                           
##                   Kappa : 0.9576          
##                                           
##  Mcnemar's Test P-Value : 0.4795          
##                                           
##             Sensitivity : 0.9444          
##             Specificity : 1.0000          
##          Pos Pred Value : 1.0000          
##          Neg Pred Value : 0.9726          
##              Prevalence : 0.3364          
##          Detection Rate : 0.3178          
##    Detection Prevalence : 0.3178          
##       Balanced Accuracy : 0.9722          
##                                           
##        'Positive' Class : No              
## 

mcrp6

## Confusion Matrix and Statistics
## 
##           Reference
## Prediction No Yes
##        No   3   3
##        Yes  6  14
##                                           
##                Accuracy : 0.6538          
##                  95% CI : (0.4433, 0.8279)
##     No Information Rate : 0.6538          
##     P-Value [Acc > NIR] : 0.5894          
##                                           
##                   Kappa : 0.1702          
##                                           
##  Mcnemar's Test P-Value : 0.5050          
##                                           
##             Sensitivity : 0.3333          
##             Specificity : 0.8235          
##          Pos Pred Value : 0.5000          
##          Neg Pred Value : 0.7000          
##              Prevalence : 0.3462          
##          Detection Rate : 0.1154          
##    Detection Prevalence : 0.2308          
##       Balanced Accuracy : 0.5784          
##                                           
##        'Positive' Class : No              
## 

Tabla de Resultados

resultados <- data.frame(
  svmLinear = c(as.numeric(mcre1$overall["Accuracy"]), as.numeric(mcrp1$overall["Accuracy"])),
  svmRadial = c(as.numeric(mcre2$overall["Accuracy"]), as.numeric(mcrp2$overall["Accuracy"])),
  svmPoly   = c(as.numeric(mcre3$overall["Accuracy"]), as.numeric(mcrp3$overall["Accuracy"])),
  rpart     = c(as.numeric(mcre4$overall["Accuracy"]), as.numeric(mcrp4$overall["Accuracy"])),
  nnet      = c(as.numeric(mcre5$overall["Accuracy"]), as.numeric(mcrp5$overall["Accuracy"])),
  rf        = c(as.numeric(mcre6$overall["Accuracy"]), as.numeric(mcrp6$overall["Accuracy"]))
)

rownames(resultados) <- c("Precisión de entrenamiento", "Precisión de prueba")
resultados

##                            svmLinear svmRadial   svmPoly     rpart      nnet
## Precisión de entrenamiento 0.9065421 0.9813084 0.9813084 0.8037383 0.9813084
## Precisión de prueba        0.5384615 0.6538462 0.5384615 0.5769231 0.4230769
##                                   rf
## Precisión de entrenamiento 0.9813084
## Precisión de prueba        0.6538462

# Ordenado por precisión de prueba
resultados_ordenados <- resultados[, order(as.numeric(resultados["Precisión de prueba", ]), decreasing = TRUE)]
resultados_ordenados

##                            svmRadial        rf     rpart svmLinear   svmPoly
## Precisión de entrenamiento 0.9813084 0.9813084 0.8037383 0.9065421 0.9813084
## Precisión de prueba        0.6538462 0.6538462 0.5769231 0.5384615 0.5384615
##                                 nnet
## Precisión de entrenamiento 0.9813084
## Precisión de prueba        0.4230769

Conclusiones

# Top model por accuracy de prueba
mejor_modelo <- names(which.max(resultados["Precisión de prueba", ]))
mejor_acc <- max(resultados["Precisión de prueba", ])

mejor_modelo

## [1] "svmRadial"

mejor_acc

## [1] 0.6538462

En este análisis se entrenaron 6 modelos de clasificación para predecir m1_purchase usando un split 80-20 y validación cruzada 10-fold.
Se comparó el desempeño con Accuracy y matrices de confusión en entrenamiento y prueba.
El mejor modelo (según Accuracy de prueba) fue: svmRadial con un Accuracy de 0.65.
Para elegir el modelo final, además de la precisión, se revisa que la diferencia entre entrenamiento y prueba no sea grande, para evitar sobreajuste.