deep learning model

1 files changed, 128 insertions, 0 deletions

diff --git a/models.R b/models.R
new file mode 100644
index 0000000..1444a8e
--- /dev/null
+++ b/models.R
@@ -0,0 +1,128 @@
+# Import necessary libraries
+library(data.table)
+library(caret)
+library(h2o)
+localH2O = h2o.init()
+
+# Importing the Network Intrusion Data set
+dataset <- fread("2020.10.01.csv")
+dataset = na.omit(dataset)
+dataset <- dataset[, -c(12, 13)]
+
+# Encoding 'label' as Numeric Variable
+dataset$label <- factor(dataset$label,
+                           levels = c("benign", "malicious", "outlier"),
+                           labels = c(1, 2, 3))
+dataset$label <- as.numeric(dataset$label)
+
+# Remove Redundant Features - First Find Correlated Features
+correlationMatrix <- cor(dataset)
+highlyCorrelated <- findCorrelation(correlationMatrix, cutoff=0.5)
+print(highlyCorrelated)
+
+df <- dataset[, c(8,2,7,3,5,12,13)]
+df <- as.h2o(df)
+
+head(dataset[, c(8,2,7,3,5,12,13)])
+
+
+# set the predictor and response columns
+predictors <- c("num_pkts_in", "bytes_in", "num_pkts_out", "bytes_out",
+                "dest_port", "total_entropy")
+response <- "label"
+
+# split the dataset into train and test sets
+df_splits <- h2o.splitFrame(data =  df, ratios = 0.8)
+train <- df_splits[[1]]
+test <- df_splits[[2]]
+
+
+# Build and train the model:
+dl <- h2o.deeplearning(x = 1:6,
+                       y = "label",
+                       distribution = "tweedie",
+                       hidden = c(1),
+                       epochs = 1000,
+                       train_samples_per_iteration = -1,
+                       reproducible = TRUE,
+                       activation = "Tanh",
+                       single_node_mode = FALSE,
+                       balance_classes = FALSE,
+                       force_load_balance = FALSE,
+                       seed = 23123,
+                       tweedie_power = 1.5,
+                       score_training_samples = 0,
+                       score_validation_samples = 0,
+                       training_frame = df,
+                       stopping_rounds = 0)
+
+# Eval performance:
+perf <- h2o.performance(dl)
+perf
+
+# Generate predictions on a test set (if necessary):
+pred <- h2o.predict(dl, newdata = df)
+pred
+summary(dl)
+plot(dl)
+
+# Save the model
+dl_model <- h2o.saveModel(object = dl, 
+                            path = "/Users/lucifer/Documents/projects/NetworkIntrusionDetection/models", 
+                          force = TRUE)
+print(dl_model)
+
+h2o.varimp_plot(dl)
+h2o.learning_curve_plot(dl)
+
+
+
+
+
+
+
+
+ind <- createDataPartition(dataset$label, p=0.6, list=FALSE)
+dataset.train <- dataset[ind,]
+dataset.test <- dataset[-ind,]
+
+
+
+
+
+
+
+# Decision Tree
+tree <- rpart(label ~., data = dataset.train)
+rpart.plot(tree)
+printcp(tree)
+plotcp(tree)
+p <- predict(tree, dataset.train)
+confusionMatrix(p, dataset.train$label, positive='y')
+
+
+
+
+# Split the class attribute
+dataset.traintarget <- dataset[ind == 1, 5]
+dataset.testtarget <- dataset[ind==2, 5]
+
+
+# Remove Redundant Features - First Find Correlated Features
+correlationMatrix <- cor(dataset)
+highlyCorrelated <- findCorrelation(correlationMatrix, cutoff=0.5)
+print(highlyCorrelated)
+
+dataset <- dataset[, c(8,2,7,3,5,12,13)]
+
+
+
+
+
+
+
+
+
+
+
+