lab4

1. Há desbalanceamento das classes (isto é, uma classe tem muito mais instâncias que outra)? Em que proporção? Quais efeitos colaterais o desbalanceamento de classes pode causar no classificador?

classes <- treino %>% count(situacao_final)
classes

## # A tibble: 2 x 2
##   situacao_final     n
##   <fct>          <int>
## 1 eleito           416
## 2 nao_eleito      3719

Há um desbalanceamento já que o número de observações não é o mesmo para todas as unidades de análise, quando a classe de interesse é mais rara que a classe majoritária. Quando as classe são desbalanceadas a acurácia não é boa métrica e o classificar tende a prever a classe majoritária.

2. Treine: um modelo de regressão logística, uma árvore de decisão e um modelo de adaboost. Tune esses modelos usando validação cruzada e controle overfitting se necessário, considerando as particularidades de cada modelo.

Para balancear o modelo usamos under sampling, onde selecionamos aleatoriamente um subconjunto de amostras da classe com mais instâncias para corresponder ao número de amostras provenientes de cada classe.

treino2 <- treino %>%
  select(-nome)

dataPartition <- createDataPartition(y = treino2$situacao_final, p=0.75, list=FALSE)
treino3 <- treino2[ dataPartition, ]

ctrl <- trainControl(method = "repeatedcv", 
                     repeats = 5,
                     verboseIter = FALSE,
                     sampling = "down")

arvore1 <- train(situacao_final ~ .,
                 data = treino3,
                 method = "rpart",
                 cp=0.001, 
                 maxdepth=20,
                 trControl = ctrl)
arvore1

## CART 
## 
## 3102 samples
##   26 predictor
##    2 classes: 'eleito', 'nao_eleito' 
## 
## No pre-processing
## Resampling: Cross-Validated (10 fold, repeated 5 times) 
## Summary of sample sizes: 2792, 2792, 2792, 2792, 2792, 2791, ... 
## Addtional sampling using down-sampling
## 
## Resampling results across tuning parameters:
## 
##   cp          Accuracy   Kappa    
##   0.02243590  0.8864603  0.5727641
##   0.07211538  0.8864603  0.5727641
##   0.31089744  0.8864603  0.5727641
## 
## Accuracy was used to select the optimal model using the largest value.
## The final value used for the model was cp = 0.3108974.

Observação: Não consegui tratar o erro (glm.fit: fitted probabilities numerically 0 or 1 occurredprediction from a rank-deficient fit may be misleadingprediction from a rank-deficient fit may be misleadingprediction from a rank-deficient fit may be misleading), mas o modelo foi gerado.

formula = as.formula(situacao_final ~ .)
modeloreglog <- train(formula,
                  data = treino3,
                  method="glm",
                  family="binomial",
                  trControl = ctrl,
                  na.action = na.omit)

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

## Warning: glm.fit: algorithm did not converge

## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

modeloreglog

## Generalized Linear Model 
## 
## 3102 samples
##   26 predictor
##    2 classes: 'eleito', 'nao_eleito' 
## 
## No pre-processing
## Resampling: Cross-Validated (10 fold, repeated 5 times) 
## Summary of sample sizes: 2792, 2792, 2791, 2792, 2792, 2792, ... 
## Addtional sampling using down-sampling
## 
## Resampling results:
## 
##   Accuracy   Kappa    
##   0.7658264  0.3147278

#modeloada <- train(formula,
#                data=treino3,
#                trControl = ctrl,
#                method = "adaboost")
#modeloada

3. Reporte acurácia, precision, recall e f-measure no treino e validação. Como você avalia os resultados? Justifique sua resposta.

A acurácia é a medida de desempenho mais intuitiva e é representa a proporção de observação corretamente previstas para as observações totais. Já a precisão é a proporção de observações positivas corretamente previstas para as observações positivas previstas. Alta precisão está relacionada à baixa taxa de falso positivo. Recall é a proporção de observações positivas corretamente previstas para todas as observações na classe real. Por fim, o f1 é a média ponderada de precisão e recall. Portanto, essa pontuação leva em conta falsos positivos e falsos negativos. O f1 é mais útil do que acurácia caso a se tenha uma distribuição de classe desigual. A acurácia funciona melhor se falsos positivos e falsos negativos têm custo semelhante. Se o custo de falsos positivos e falsos negativos for muito diferente, é melhor olhar precisão e recall.

Como os dados foram balanceados, consideramos que a acurácia é uma boa métrica, bem como o f1. Sendo assim, os resultados obtidos com a árvore de decisão foram levemente melhores que os demais.

treino3$predictions_arvore <- predict(arvore1, newdata = treino3)

TP <- treino3 %>% filter(situacao_final == "eleito", predictions_arvore == "eleito") %>% nrow()
TN <- treino3 %>% filter(situacao_final == "nao_eleito" , predictions_arvore == "nao_eleito" ) %>% nrow()
FP <- treino3 %>% filter(situacao_final == "nao_eleito" , predictions_arvore == "eleito") %>% nrow() 
FN <- treino3 %>% filter(situacao_final == "eleito", predictions_arvore == "nao_eleito" ) %>% nrow()
accuracy <- (TP + TN)/(TP + TN + FP + FN) 
precision <- TP / (TP + FP)
recall <- TP / (TP + FN)
f1 <- 2*(recall * precision) / (recall + precision)

accuracy

## [1] 0.8668601

precision

## [1] 0.4289733

recall

## [1] 0.9775641

f1

## [1] 0.5962854

confusionMatrix(treino3$predictions_arvore, treino3$situacao_final)

## Confusion Matrix and Statistics
## 
##             Reference
## Prediction   eleito nao_eleito
##   eleito        305        406
##   nao_eleito      7       2384
##                                           
##                Accuracy : 0.8669          
##                  95% CI : (0.8544, 0.8786)
##     No Information Rate : 0.8994          
##     P-Value [Acc > NIR] : 1               
##                                           
##                   Kappa : 0.5307          
##  Mcnemar's Test P-Value : <2e-16          
##                                           
##             Sensitivity : 0.97756         
##             Specificity : 0.85448         
##          Pos Pred Value : 0.42897         
##          Neg Pred Value : 0.99707         
##              Prevalence : 0.10058         
##          Detection Rate : 0.09832         
##    Detection Prevalence : 0.22921         
##       Balanced Accuracy : 0.91602         
##                                           
##        'Positive' Class : eleito          
## 

treino3$predictions_modeloreglog <- predict(modeloreglog, newdata = treino3)

## Warning in predict.lm(object, newdata, se.fit, scale = 1, type =
## ifelse(type == : prediction from a rank-deficient fit may be misleading

TP <- treino3 %>% filter(situacao_final == "eleito", predictions_modeloreglog == "eleito") %>% nrow()
TN <- treino3 %>% filter(situacao_final == "nao_eleito" , predictions_modeloreglog == "nao_eleito" ) %>% nrow()
FP <- treino3 %>% filter(situacao_final == "nao_eleito" , predictions_modeloreglog == "eleito") %>% nrow() 
FN <- treino3 %>% filter(situacao_final == "eleito", predictions_modeloreglog == "nao_eleito" ) %>% nrow()
accuracy <- (TP + TN)/(TP + TN + FP + FN) 
precision <- TP / (TP + FP)
recall <- TP / (TP + FN)
f1 <- 2*(recall * precision) / (recall + precision)

accuracy

## [1] 0.7814313

precision

## [1] 0.3151515

recall

## [1] 1

f1

## [1] 0.4792627

confusionMatrix(treino3$predictions_modeloreglog, treino3$situacao_final)

## Confusion Matrix and Statistics
## 
##             Reference
## Prediction   eleito nao_eleito
##   eleito        312        678
##   nao_eleito      0       2112
##                                           
##                Accuracy : 0.7814          
##                  95% CI : (0.7665, 0.7959)
##     No Information Rate : 0.8994          
##     P-Value [Acc > NIR] : 1               
##                                           
##                   Kappa : 0.3852          
##  Mcnemar's Test P-Value : <2e-16          
##                                           
##             Sensitivity : 1.0000          
##             Specificity : 0.7570          
##          Pos Pred Value : 0.3152          
##          Neg Pred Value : 1.0000          
##              Prevalence : 0.1006          
##          Detection Rate : 0.1006          
##    Detection Prevalence : 0.3191          
##       Balanced Accuracy : 0.8785          
##                                           
##        'Positive' Class : eleito          
## 

#treino3$predictions_modeloada <- predict(modeloada, newdata = treino3)

#TP <- treino3 %>% filter(situacao_final == "eleito", predictions_modeloada == "eleito") %>% nrow()
#TN <- treino3 %>% filter(situacao_final == "nao_eleito" , predictions_modeloada == "nao_eleito" ) %>% nrow()
#FP <- treino3 %>% filter(situacao_final == "nao_eleito" , predictions_modeloada == "eleito") %>% nrow() 
#FN <- treino3 %>% filter(situacao_final == "eleito", predictions_modeloada == "nao_eleito" ) %>% nrow()
#accuracy <- (TP + TN)/(TP + TN + FP + FN) 
#precision <- TP / (TP + FP)
#recall <- TP / (TP + FN)
#f1 <- 2*(recall * precision) / (recall + precision)

#accuracy
#precision
#recall
#f1

#confusionMatrix(treino3$predictions_modeloada, treino3$situacao_final)

4. Interprete as saídas dos modelos. Quais atributos parecem ser mais importantes de acordo com cada modelo? Crie pelo menos um novo atributo que não está nos dados originais e estude o impacto desse atributo.

Segundo a árvore de decisão os atributos mais importantes são total_despesa e total_receita.

varImp(arvore1)

## rpart variable importance
## 
##   only 20 most important variables shown (out of 384)
## 
##                                                                                                                                         Overall
## total_despesa                                                                                                                            100.00
## total_receita                                                                                                                            100.00
## quantidade_despesas                                                                                                                       86.78
## quantidade_fornecedores                                                                                                                   82.18
## recursos_de_pessoas_juridicas                                                                                                             78.22
## `setor_economico_receitaPreparação de documentos e serviços especializados de apoio administrativo não especificados anteriormente`    0.00
## `setor_economico_despesaComércio varejista especializado de eletrodomésticos e equipamentos de áudio e vídeo`                          0.00
## descricao_ocupacaoFAXINEIRO                                                                                                                0.00
## descricao_ocupacaoPEDAGOGO                                                                                                                 0.00
## partidoPP                                                                                                                                  0.00
## quantidade_doacoes                                                                                                                         0.00
## descricao_ocupacaoMASSAGISTA                                                                                                               0.00
## `descricao_ocupacaoPROFESSOR DE ENSINO MÉDIO`                                                                                             0.00
## `setor_economico_despesaPromoção de vendas`                                                                                              0.00
## `setor_economico_receitaFabricação de esquadrias de madeira e de peças de madeira para instalações industriais e comerciais`          0.00
## descricao_ocupacaoDESENHISTA                                                                                                               0.00
## `setor_economico_receitaHotéis`                                                                                                           0.00
## partidoPCO                                                                                                                                 0.00
## `setor_economico_despesaServiços de pré-impressão`                                                                                      0.00
## `grauLÊ E ESCREVE`                                                                                                                        0.00

varImp(modeloreglog)

## glm variable importance
## 
##   only 20 most important variables shown (out of 198)
## 
##                                                                                                            Overall
## `setor_economico_despesaEdição integrada à impressão de cadastros, listas e outros produtos gráficos`  100.00
## `setor_economico_receitaAtividades de organizações políticas`                                             87.32
## quantidade_doadores                                                                                          78.84
## sexoMASCULINO                                                                                                78.57
## UFAL                                                                                                         64.60
## UFAP                                                                                                         64.60
## UFCE                                                                                                         64.60
## UFBA                                                                                                         64.60
## UFMA                                                                                                         64.60
## UFPB                                                                                                         64.60
## UFES                                                                                                         64.60
## UFMT                                                                                                         64.60
## UFMG                                                                                                         64.60
## UFRR                                                                                                         64.60
## UFRO                                                                                                         64.60
## UFPE                                                                                                         64.60
## UFPA                                                                                                         64.60
## UFRN                                                                                                         64.60
## UFDF                                                                                                         64.60
## UFSC                                                                                                         64.60

#varImp(modeloada)

5. Envie seus melhores modelos à competição do Kaggle.

Para o modelo de submissão foram utilizadas as variáveis de maior importância.

selecionadados <- treino3 %>% select(total_receita, total_despesa, situacao_final, quantidade_despesas, quantidade_doadores, recursos_de_pessoas_juridicas)

arvore <- train(situacao_final ~ .,
                 data = selecionadados,
                 method = "rpart",
                 cp=0.001, 
                 maxdepth=20,
                 trControl = ctrl)

predicao <- predict(arvore, teste)
teste$prediction <- predicao
predicaofinal <- teste %>% select(ID, prediction)

write.csv(predicaofinal, file = "lab4_finalpredicao.csv", row.names = FALSE)