PSF : R Package for Pattern Sequence based Forecasting Algorithm

Install library

Download the Package and install with instruction:

library(PSF)

pred_for_w (data_in, w, k, next_val)

This Function predict the values for given data, Window size (W) and Cluster size (k).

• data_in as Input data, in any format (data matrix data frame or vector). All variables should be numeric and NA values will get removed while execution.
• w as window size (Can be obtained with function optimum_w)
• k as cluster size for Kmeans (Can be obtained with function optimum_k)
• next_val as Integer number. It states the number of predicted values to be obtained.

# Considering `data_in` = iris[1], `w` = 3, `k` = 4 and `next_val` = 5
pred_for_w(iris[1], 3, 4, 5)

## [1] 7.082143 6.271795 5.623684 5.623684 5.623684

This function returns the predicted values and corresponding graph plot. The graph region with Red Color is the original data, whereas the region with Blue color is corresponding to the predicted values.

pred_for_w_plot (data_in, w, k, next_val)

This Function is similar to pred_for_w(), except this function is able to plot the predicted values.

# Considering `data_in` = iris[1], `w` = 3, `k` = 4 and `next_val` = 5
pred_for_w_plot(iris[1], 3, 4, 5)

## $Predicted_Values
## [1] 7.082143 6.271795 5.623684 5.623684 5.623684
## 
## $Plot
## NULL

optimum_w (data_in, next_val)

This Function calculates the optimum value of Window size w to be used in prediction function such that the RMSE and MAE values should be minimum.

• data_in as Input data, in any format (data matrix data frame or vector). All variables should be numeric and NA values will get removed while execution.
• next_val as Integer number. Such that last next_val numbers will be used for test of prediction accuracy.

# Considering `data_in` = iris[1] and `next_val` = 5
optimum_w(iris[1], 5)

## $Optimum_W
## [1] 1
## 
## $RMSE_Values
## [1] 0.3059953
## 
## $Prediction
##   Original_Data W_size.1
## 1           6.7 6.461538
## 2           6.3 6.461538
## 3           6.5 6.461538
## 4           6.2 6.461538
## 5           5.9 6.461538

This Function returns the optimum value for w, it’s corresponding RMSE values, prediction values and plots. The predicted values in optimum_w() function are different than those obtained in pred_for_w(), since, optimum_w() function removes last next_val numbers of integers and predicts that much values and compare them with original data.

optimum_k (data_in)

This Function is to determine the Optimum value of Cluster size (K) based on Average silhouette method.

• data_in as Input data, in any format (data matrix data frame or vector). All variables should be numeric and NA values will get removed while execution.

# Considering `data_in` = iris[1] 
optimum_k(iris[1])

## [1] 10

This Function returns K value, an integer as optimum number clusters.

AUTO_PSF (data_in, next_val)

This Function takes input data and autogenerate optimum Window size (W) and cluster size (K) and predicts the next_val numbers of future values

• data_in as Input data, in any format (data matrix data frame or vector). All variables should be numeric and NA values will get removed while execution.
• next_val as Integer number. It states the number of predicted values to be obtained.

# Considering `data_in` = iris[1] and next_val = 3
AUTO_PSF(iris[1],3)

## $Predicted_Values
## [1] 5.546154 5.546154 5.772222

This function calculate suitable values of w and k such that mean erroe between original and predicted data should be minimum. And ultimately, forecasts the series with PSF algorithm methodology.