trainData <- read.csv("C:/Users/TiagoMonteiro/Downloads/CDados/CDados/dataset1/aps_failure_training_set.csv", na.strings="na")
testData <- read.csv("C:/Users/TiagoMonteiro/Downloads/CDados/CDados/dataset1/aps_failure_test_set.csv", na.strings="na")

qplot(as.factor(trainData$class), xlab = "class")
qplot(as.factor(testData$class), xlab = "class")


trainData <- trainData[, -which(colMeans(is.na(trainData)) > 0.7)]
testData <- testData[, -which(colMeans(is.na(testData)) > 0.7)]

#impute test data
preProcValues <- preProcess(testData, method = c("knnImpute"), k = 3)

# use it to predict all the missing values
testData_knnimp <- predict(preProcValues, testData)

set.seed(1234)

#clustering
cl <- makeCluster(8, type="SOCK")
registerDoSNOW(cl)


#omit
trainData_omit <- na.omit(trainData);

#knn impute of train
trainData_imputation <- preProcess(trainData, method = c("knnImpute"), k = 3)

# use it to predict all the missing values
trainData_imputation <- predict(trainData_imputation, trainData)

#mean
trainData_mean <- mice(trainData, m=1, maxit = 5, method = "mean", seed = 500)
testData_mean <- mice(testData, m=1, maxit = 5, method = "mean", seed = 500)

#mean s/ constants linears

attrToDelete <- subset(trainData_mean$loggedEvents, meth == "collinear" | meth == "constant" )
attrToDeleteNames <- as.character (attrToDelete [, "out"])

trainData_mean <- complete(trainData_mean, 1)
trainData_mean_clean <- trainData_mean[, !(names(trainData_mean) %in% attrToDeleteNames)]

attrToDelete <- subset(testData_mean$loggedEvents, meth == "collinear" | meth == "constant" )
attrToDeleteNames <- as.character (attrToDelete [, "out"])

testData_mean <- complete(testData_mean, 1)
testData_mean_clean <- testData_mean[, !(names(testData_mean) %in% attrToDeleteNames)]


#feature selection

trainData_mean_clean_feature_selection <- trainData_mean_clean

trainData_mean_clean_feature_selection <- gain.ratio( class~., trainData_mean_clean_feature_selection)
trainData_mean_clean_feature_selection <- t(trainData_mean_clean_feature_selection)
trainData_mean_clean_feature_selection <- trainData_mean_clean_feature_selection[which(!colSums(trainData_mean_clean_feature_selection) == 0)]



#KNN
grid <- expand.grid(k = c((seq(3, 17, by=2))))


 testData1$as_000 <- NULL
 trainData_omit$cd_000 <- NULL

results.model_knn_omit_pp_norm = train(trainData_omit[-1], trainData_omit$class, method="knn", tuneGrid=grid, preProcess=c("center", "scale"))
results.model_knn_omit_pp_norm_pca = train(trainData_omit[-1], trainData_omit$class, method="knn", tuneGrid=grid, preProcess=c("center", "scale", "pca"))

results.model_knn_clean_pp_norm = train(trainData_mean_clean[-1], trainData_mean_clean$class, method="knn", tuneGrid=grid, preProcess=c("center", "scale"))
results.model_knn_clean_pp_norm_pca = train(trainData_mean_clean[-1], trainData_mean_clean$class, method="knn", tuneGrid=grid, preProcess=c("center", "scale", "pca"))


#NB

grid <- data.frame(fL=c(0.5,1.0), usekernel = TRUE, adjust=c(0.5,1.0))


results.model_nb_omit_pp_norm = train(trainData_omit[-1], trainData_omit$class, method="nb", tuneGrid=grid, preProcess=c("center", "scale"))
results.model_nb_omit_pp_norm_pca = train(trainData_omit[-1], trainData_omit$class, method="nb", tuneGrid=grid, preProcess=c("center", "scale", "pca"))

results.model_nb_mean_pp_norm = train(trainData_mean[-1], trainData_mean$class, method="nb", tuneGrid=grid, preProcess=c("center", "scale"))
results.model_nb_mean_pp_norm_pca = train(trainData_mean[-1], trainData_mean$class, method="nb", tuneGrid=grid, preProcess=c("center", "scale", "pca"))

results.model_nb_clean_pp_norm = train(trainData_mean_clean[-1], trainData_mean_clean$class, method="nb", tuneGrid=grid, preProcess=c("center", "scale"))
results.model_nb_clean_pp_norm_pca = train(trainData_mean_clean[-1], trainData_mean_clean$class, method="nb", tuneGrid=grid, preProcess=c("center", "scale", "pca"))


model_nb <- train(class ~ .,data = trainData, method = "nb",tuneList = NULL, preProcess = c("center", "scale"))

preds <- predict(model_nb, testData)
confusionMatrix(preds, testData$class)

#KNN

grid <- expand.grid(k = c((seq(3, 17, by=2))))
model_cv <- train(class ~ ., data = trainData, method="knn", tuneGrid=grid, preProcess = c("pca"))

#J48
library("RWeka")
truck_j48 = J48(trainData$class ~ ., data = trainData)
plot(truck_j48)

eval_j48 <- evaluate_Weka_classifier(truck_j48, numFolds = 10, complexity = FALSE,
                                     seed = 1, class = TRUE)
eval_j48
preds <- predict(truck_j48, newdata = testData)
confusionMatrix(preds, testData$class)

#randomForest
library("randomForest")
set.seed(1234)

random_forest <- randomForest(trainData$class~., data=trainData, controls=cforest_unbiased(ntree=2000, mtry=3))
preds <- predict(random_forest, newdata = testData)
confusionMatrix(preds, testData$class)