Trains and tests SVM on two features (relative strength index and trend over x observations)

svm_rsi_trend.R

# Some code to asses an SVM with a two-dimensional feature space:
# trend (price - simple moving average) and relative strength index.
# The code is an adaptation of the code found in the following linkedin post:
# "Trading the RSI using a Support Vector Machine"
# "https://www.linkedin.com/pulse/article/20141103165037-172934333-trading-the-rsi-using-a-support-vector-machine"

# Settings
sma.window = 50 # Number of observations in simple moving average.
rsi.window = 3  # Number of observation in relative strength index (RSI)

# Specify and load required R packages. 
packs = c(
  "quantmod",   # quant trading
  "e1071",      # machine learning library containing SVM
  "ggplot2",    # plotting
  "xts"         # working with time-series
)

sapply(packs, require, character.only=TRUE)   # Load the packages.


# Load the data (from the working directory).
file.name = "AUDUSD.csv"
dat = read.csv(file.name, stringsAsFactors=FALSE)

# Convert the data to an xts time series object.
dat = as.xts(
  dat[ , -1],
  order.by= as.POSIXct(dat[ , 1], tz="GMT", format="%m/%d/%y %H:%M")
)


# Create the data set for training, testing and validation.
dat = data.frame(
  "rsi" = unname(RSI(dat$Open, n=rsi.window)),  # relative strength indicator
  "trend" = unname((dat$Open - SMA(dat$Open, n=sma.window))),
  "direction" = unname(ifelse(dat$Close > dat$Open, "UP", "DOWN"))
)

dat = dat[complete.cases(dat), ]   # Remove rows containing NAs.

# Separate the data into 60% training set to build our model, 20% test set to test the patterns we found, and 20% validation set to run our strategy over new data

Training = dat[1:4528,]
Test = dat[4529:6038,]
Val = dat[6039:7548,]


# Train support vector machine.
machine = svm(
  direction ~ rsi + trend,
  data=Training,
  kernel="polynomial",
  cost=1,
  gamma=1/2
)


# Generate predictions for the training, test, and validation sets.
TrainingPredictions = predict(machine, Training, type="class")
TestPredictions = predict(machine, Test, type="class")
ValPredictions = predict(machine, Val, type="class")


ShowContours = function() {
# Plot the predictions distribution across the feature space using the test set.
  ggplot(
    data.frame(Training, TrainingPredictions),
    aes(x=trend,y=rsi)
  ) + stat_density2d(
     geom="contour",
     aes(color=TrainingPredictions)
  ) + labs(
     title= "Polynomial kernel",
     x="Trend",
     y="RSI",
     color="Training Predictions"
  )
}


HitRatios = function() {
  # Calculates the fraction of test cases where the classification matches
  # the outcome.
  print(mean(TrainingPredictions == Training$direction) * 100)
  print(mean(TestPredictions == Test$direction) * 100)
  print(mean(ValPredictions == Val$direction) * 100)
}