library(caret)

library(rpart)
library(rpart.plot)

library(tidyverse)

library(lubridate)
# Set seed untuk reproducibility
set.seed(123)

# Membuat data time series
create_ts_data <- function(n_days = 365*3) {
  # Tanggal dari 1 Jan 2020 hingga 3 tahun ke depan
  dates <- seq(as.Date("2020-01-01"), by = "day", length.out = n_days)
  
  # Komponen tren (meningkat seiring waktu)
  trend <- seq(1, n_days) * 0.05
  
  # Komponen musiman (pola tahunan)
  seasonal <- 15 * sin(2 * pi * seq(1, n_days)/365) + 
              5 * sin(2 * pi * seq(1, n_days)/7)  # pola mingguan
  
  # Noise random
  noise <- rnorm(n_days, 0, 5)
  
  # Gabungkan komponen
  values <- trend + seasonal + noise
  
  # Buat data frame
  ts_data <- data.frame(
    date = dates,
    value = values,
    # Fitur tambahan yang mungkin berguna untuk model
    day_of_week = wday(dates),
    month = month(dates),
    is_weekend = ifelse(wday(dates) %in% c(1, 7), 1, 0),
    day_of_year = yday(dates),
    year = year(dates)
  )
  
  # Buat lag features (nilai pada hari-hari sebelumnya)
  for (i in 1:7) {
    ts_data[paste0("lag_", i)] <- c(rep(NA, i), head(ts_data$value, -i))
  }
  
  # Hapus baris dengan NA
  ts_data <- ts_data[8:nrow(ts_data),]
  
  return(ts_data)
}

# Buat data
ts_data <- create_ts_data()

# Plot data
ggplot(ts_data, aes(x = date, y = value)) +
  geom_line() +
  labs(title = "Data Time Series Simulasi", 
       x = "Tanggal", 
       y = "Nilai") +
  theme_minimal()

# Target variable: nilai pada hari berikutnya
ts_data$target <- c(ts_data$value[-1], NA)
ts_data <- ts_data[-nrow(ts_data),]  # Hapus baris terakhir karena target NA

# Tampilkan beberapa baris data
head(ts_data)

# Set seed untuk reproducibility
set.seed(456)

# Membuat data klasifikasi
create_classification_data <- function(n_samples = 1000) {
  # Buat fitur
  x1 <- rnorm(n_samples, mean = 0, sd = 1)
  x2 <- rnorm(n_samples, mean = 0, sd = 1)
  x3 <- rnorm(n_samples, mean = 0, sd = 1)
  x4 <- rnorm(n_samples, mean = 0, sd = 1)
  
  # Feature dengan sedikit noise
  x5 <- runif(n_samples, min = -2, max = 2)
  
  # Feature kategoris
  x6 <- sample(c("A", "B", "C"), n_samples, replace = TRUE)
  
  # Buat aturan untuk kelas
  # Kelas 1: x1 > 0.5 dan x2 > 0
  # Kelas 2: x1 < -0.5 dan x2 < 0
  # Kelas 3: lainnya
  class <- rep("Class_3", n_samples)
  class[x1 > 0.5 & x2 > 0] <- "Class_1"
  class[x1 < -0.5 & x2 < 0] <- "Class_2"
  
  # Tambah noise pada kelas (5% data diacak)
  noise_idx <- sample(1:n_samples, round(0.05 * n_samples))
  class[noise_idx] <- sample(c("Class_1", "Class_2", "Class_3"), length(noise_idx), replace = TRUE)
  
  # Buat dataframe
  class_data <- data.frame(
    x1 = x1,
    x2 = x2,
    x3 = x3,
    x4 = x4,
    x5 = x5,
    x6 = x6,
    target = factor(class)
  )
  
  return(class_data)
}

# Buat data
class_data <- create_classification_data()

# Plot data untuk visualisasi
ggplot(class_data, aes(x = x1, y = x2, color = target)) +
  geom_point(alpha = 0.6) +
  labs(title = "Data Klasifikasi Simulasi", 
       x = "Fitur x1", 
       y = "Fitur x2") +
  theme_minimal()

library(MLmetrics)

# Split data klasifikasi
set.seed(789)
train_idx <- createDataPartition(class_data$target, p = 0.7, list = FALSE)
class_train <- class_data[train_idx, ]
class_test <- class_data[-train_idx, ]

# Definisi training control
class_ctrl <- trainControl(
  method = "cv",
  number = 10,
  verboseIter = TRUE,
  classProbs = TRUE,
  summaryFunction = multiClassSummary  # untuk metrik multiclass
)

# Train model decision tree
class_dt_model <- train(
  target ~ .,
  data = class_train,
  method = "rpart",
  trControl = class_ctrl,
  tuneLength = 10,
  metric = "Accuracy"
)

# Plot hasil tuning
plot(class_dt_model)

# Visualisasi pohon keputusan
rpart.plot(class_dt_model$finalModel, 
           extra = 101, 
           box.palette = "auto", 
           shadow.col = "gray")

# Prediksi pada data testing
class_predictions <- predict(class_dt_model, newdata = class_test)
class_probs <- predict(class_dt_model, newdata = class_test, type = "prob")

# Evaluasi model
conf_matrix <- confusionMatrix(class_predictions, class_test$target)
print(conf_matrix)

# Plot confusion matrix
library(viridis)

conf_data <- as.data.frame(conf_matrix$table)
conf_data$Prediction <- factor(conf_data$Prediction, levels = rev(levels(conf_data$Prediction)))

ggplot(conf_data, aes(x = Reference, y = Prediction, fill = Freq)) +
  geom_tile() +
  geom_text(aes(label = Freq), color = "white", size = 4) +
  scale_fill_viridis() +
  labs(title = "Confusion Matrix", x = "Actual", y = "Predicted") +
  theme_minimal()

# Visualisasi batas keputusan
# Buat grid untuk prediksi
x1_range <- seq(min(class_data$x1) - 0.5, max(class_data$x1) + 0.5, length.out = 100)
x2_range <- seq(min(class_data$x2) - 0.5, max(class_data$x2) + 0.5, length.out = 100)
grid <- expand.grid(x1 = x1_range, x2 = x2_range)

# Tetapkan nilai untuk fitur lain (ambil nilai rata-rata/modus)
grid$x3 <- mean(class_data$x3)
grid$x4 <- mean(class_data$x4)
grid$x5 <- mean(class_data$x5)
grid$x6 <- names(which.max(table(class_data$x6)))

# Prediksi pada grid
grid_pred <- predict(class_dt_model, newdata = grid)

# Plot batas keputusan
ggplot() +
  geom_tile(data = grid, aes(x = x1, y = x2, fill = grid_pred), alpha = 0.3) +
  geom_point(data = class_test, aes(x = x1, y = x2, color = target), size = 2) +
  labs(title = "Decision Tree: Batas Keputusan",
       x = "Fitur x1",
       y = "Fitur x2",
       fill = "Predicted Class",
       color = "Actual Class") +
  theme_minimal()

# Tampilkan variabel importance
var_imp <- varImp(class_dt_model)
plot(var_imp, main = "Variable Importance")