Merge pull request #1 from Randul-Malinhara/add-initial-files

Add the scripts, datasets

18098293_energyForecasting.R

@@ -0,0 +1,110 @@
+# Import libraries
+library(readxl)
+library(neuralnet)
+library(caTools)
+library(MLmetrics)
+library(Metrics)
+library(xlsx)
+
+# Build the Assumed Values and real Values table
+predictedVsreal <- function (assumed, mlp_test) {
+  assumedTest <- cbind(assumed, as.data.frame(mlp_test$net.result))
+  colnames(assumedTest) <- c("Expected Output", "NeuralNetwork Output")
+  return(assumedTest)
+}
+
+# Evaluation Function
+rmse <- function (real, predicted) 
+{
+  RMSE <- sqrt(mean((real - predicted)^2))
+  return(RMSE)
+}
+evaluationFunction <- function(real,predict) {
+  rmse_mlp <- rmse(real = real, predicted = predict)
+  mae_mlp <- Metrics::mae(real = real, predicted = predict)
+  mape_mlp <- MAPE(y_pred = predict, y_true = real)
+  return(c(rmse_mlp, mae_mlp, mape_mlp))
+}
+
+# Import Processed Data set
+powerOutage <- read_xlsx("C:/Users/RANDUL/Desktop/2nd Semester of 2nd Year/5DATA001C.2 Machine Learning and Data Mining/CW/scaled_powerusage_10.xlsx")
+
+# # For 9th Hour and 10th Hour 
+# (Note: Run the both file individually to get the input as 9th hour and 10th hour)
+# powerOutage <- read_xlsx("C:/Users/RANDUL/Desktop/2nd Semester of 2nd Year/5DATA001C.2 Machine Learning and Data Mining/CW/powerUsage-scaled-09.xlsx")
+# powerOutage <- read_xlsx("C:/Users/RANDUL/Desktop/2nd Semester of 2nd Year/5DATA001C.2 Machine Learning and Data Mining/CW/powerUsage-scaled-10.xlsx")
+
+# Split Data
+set.seed(123)
+splitRule <- sample(seq_len(nrow(powerOutage)), size = 430)
+train <- powerOutage[splitRule, ]
+test <- powerOutage[-splitRule, ]
+
+# Define X values of test data
+x_test <- test[-1]
+
+# Define Y values of test data
+y_test <- test[1]
+
+# MLP NN - Configuration 1
+relation1 <- as.formula("wineD_original_data~v2+v3+v4+v5+v6")
+mlp1 <- neuralnet(formula = relation1, data = train,hidden = c(4,3), stepmax = 1e+10)
+plot(mlp1)
+
+# Make prediction on X test data
+y_pred1 <- neuralnet::compute(mlp1, x_test)
+evaluationData1 <- predictedVsreal(y_test, y_pred1)
+mlpTestResult1 <- evaluationFunction(real = evaluationData1$`Expected Output`, predict = evaluationData1$`NeuralNetwork Output`)
+
+# MLP NN - Configuration 2
+relation2 <- as.formula("wineD_original_data~v2+v3+v4+v5+v6")
+mlp2 <- neuralnet(formula = relation2, data = train,hidden = c(4,3), stepmax = 1e+10, learningrate = 0.001)
+plot(mlp2)
+
+# Make prediction on X test data
+y_pred2 <- neuralnet::compute(mlp2, x_test)
+evaluationData2 <- predictedVsreal(y_test, y_pred2)
+mlpTestResult2 <- evaluationFunction(real = evaluationData2$`Expected Output`, predict = evaluationData2$`NeuralNetwork Output`)
+
+# MLP NN - Configuration 3
+relation3 <- as.formula("wineD_original_data~v2+v3+v4+v5+v6")
+mlp3 <- neuralnet(formula = relation3, data = train,hidden = c(3,2), stepmax = 1e+10)
+plot(mlp3)
+
+# Make prediction on X test data
+y_pred3 <- neuralnet::compute(mlp3, x_test)
+evaluationData3 <- predictedVsreal(y_test, y_pred3)
+mlpTestResult3 <- evaluationFunction(real = evaluationData3$`Expected Output`, predict = evaluationData3$`NeuralNetwork Output`)
+
+# MLP NN - Configuration 4
+relation4 <- as.formula("wineD_original_data~v2+v3+v4+v5+v6")
+mlp4 <- neuralnet(formula = relation4, data = train,hidden = c(5,2), stepmax = 1e+10, learningrate = 0.001)
+plot(mlp4)
+
+# Make prediction on X test data
+y_pred4 <- neuralnet::compute(mlp4, x_test)
+evaluationData4 <- predictedVsreal(y_test, y_pred4)
+mlpTestResult4 <- evaluationFunction(real = evaluationData4$`Expected Output`, predict = evaluationData4$`NeuralNetwork Output`)
+
+# MLP NN - Configuration 5
+train0 <- train[1:4]
+relation5 <- as.formula("wineD_original_data~v2+v3+v4")
+mlp5 <- neuralnet(formula = relation5, data = train0,hidden = c(4,3), stepmax = 1e+10, learningrate = 0.001)
+plot(mlp5)
+
+x_test0 <- test[2:4]
+
+# Make prediction on X test data
+# (Note: First try with configuration 1 and run this 98:100 and then 2, 3, 4, 5)
+y_pred5 <- neuralnet::compute(mlp5, x_test0)
+evaluationData5 <- predictedVsreal(y_test, y_pred5)
+mlpTestResult5 <- evaluationFunction(real = evaluationData5$`Expected Output`, predict = evaluationData5$`NeuralNetwork Output`)
+
+comp <- rbind(mlpTestResult1, mlpTestResult2, mlpTestResult3, mlpTestResult4, mlpTestResult5)
+colnames(comp) <- c("RMSE","MAE","MAPE")
+rownames(comp) <- c("config-1","config-2","config-3","config-4","config-5")
+comp
+
+write.xlsx(comp,
+           file = "C:/Users/RANDUL/Desktop/2nd Semester of 2nd Year/5DATA001C.2 Machine Learning and Data Mining/CW/comparison-powerusage.xlsx"
+           ,col.names = TRUE, append = TRUE, row.names = TRUE)

18098293_p_clustering.R

@@ -0,0 +1,131 @@
+# import libraries
+library(readxl)
+library(fpc)
+library(NbClust)
+library(dplyr)
+library(MASS)
+library(caret)
+library(flexclust)
+
+
+# Read the File of White wine_v2
+wineD <- read_xlsx("C:/Users/RANDUL/Desktop/2nd Semester of 2nd Year/5DATA001C.2 Machine Learning and Data Mining/CW/Whitewine_v2.xlsx")
+boxplot(wineD)
+
+# outliers removal
+wineD_summary <- summary(wineD$`residual sugar`)
+wineD_summary
+
+# Estimate interquartile range
+# (3rd interquartile minus 1st interquartile)
+interqr <- wineD_summary[[5]] - wineD_summary[[2]]
+
+
+# Identifying the bounds for outliers
+lower_limit <- wineD_summary[[2]] - (1.5 * interqr)
+upper_limit <- wineD_summary[[5]] + (1.5 * interqr)
+
+# Identifying the outliers
+outliers <- wineD %>%
+  filter(`residual sugar`> upper_limit | `residual sugar`< lower_limit)
+
+# Outliers are removed from the data frame, but a new dataframe called "no outliers" is created.
+no_outliers <- wineD %>%
+  filter(`residual sugar` < upper_limit & `residual sugar` > lower_limit)
+
+wineD_summary <- summary(wineD$`free sulfur dioxide`)
+interqr <- wineD_summary[[5]] - wineD_summary[[2]]
+
+# The bounds are established with the wineD data
+lower_limit <- wineD_summary[[2]] - (1.5 * interqr)
+upper_limit <- wineD_summary[[5]] + (1.5 * interqr)
+
+outliers <- rbind(outliers,wineD %>% 
+                    filter(`free sulfur dioxide` > upper_limit | 
+                             `free sulfur dioxide` < lower_limit))
+
+no_outliers <- no_outliers %>%
+  filter(`free sulfur dioxide` < upper_limit & `free sulfur dioxide` > lower_limit)
+
+
+
+# Repeat for fixed acidity
+wineD_summary <- summary(wineD$`total sulfur dioxide`)
+interqr <- wineD_summary[[5]] - wineD_summary[[2]]
+
+# Remember that the bounds are based on the wineD data
+lower_limit <- wineD_summary[[2]] - (1.5 * interqr)
+upper_limit <- wineD_summary[[5]] + (1.5 * interqr)
+
+# Removing fixed acidity outliers from the no_outliers data, not the wineD
+outliers <- rbind(outliers,wineD %>% 
+                    filter(`total sulfur dioxide` > upper_limit | 
+                             `total sulfur dioxide` < lower_limit))
+
+no_outliers <- no_outliers %>%
+  filter(`total sulfur dioxide` < upper_limit & `total sulfur dioxide` > lower_limit)
+
+boxplot(no_outliers)
+
+
+
+# Scaling 
+wine_stand<- scale(no_outliers[-12])
+summary(wine_stand)
+
+# funct NbClust()
+set.seed(1234)
+nc <- NbClust(wine_stand,
+              min.nc=2, max.nc=8,
+              method="kmeans")
+
+barplot(table(nc$Best.n[1,]),
+        xlab="Total Number of Clusters",
+        ylab="Total Number of Criterias",
+        main="Total Number of Clusters Chosen by 9 Criteria")
+
+ws <- 0
+for (i in 1:9){
+  ws[i] <-
+    sum(kmeans(wine_stand, centers=i)$withinss)}
+
+plot(1:9,
+     ws,
+     type="b",    
+     xlab="Total Number of Clusters",
+     ylab="Within groups sum of squares")
+
+# k means = 2
+fit.km2 <- kmeans(wine_stand,2)
+plotcluster(wine_stand, fit.km2$cluster)
+confuse <- table(no_outliers$quality,fit.km2$cluster)
+confuse
+parcoord(wine_stand, fit.km2$cluster)
+
+# k means = 3
+fit.km3 <- kmeans(wine_stand , 3)
+fit.km3
+confuse3 <- table(no_outliers$quality,fit.km3$cluster)
+confuse
+parcoord(wine_stand, fit.km2$cluster)
+
+# k means = 4
+fit.km4 <- kmeans(wine_stand, 4)
+table(no_outliers$quality,fit.km4$cluster)
+confuse
+parcoord(wine_stand, fit.km2$cluster)
+
+# k means = 5
+fit.km5 <- kmeans(wine_stand, 5)
+table(no_outliers$quality,fit.km5$cluster)
+confuse
+parcoord(wine_stand, fit.km2$cluster)
+
+plotcluster(wine_stand,fit.km5$cluster)
+
+# Evaluation with ARI for k=2 
+randIndex(confuse)
+
+# NbClust() with Manhattan distance
+clusters_manhattan <- NbClust(wine_stand,distance="manhattan",min.nc=2,max.nc=5,method="kmeans",
+                              index="all")

18098293_pre-process.R

@@ -0,0 +1,45 @@
+# Import libraries
+library(tidyverse)
+library(readxl)
+library(tidymodels)
+library(readxl)
+library(caTools)
+library(xlsx)
+
+# --------------- PREPARATION OF DATA BEFRORE APPROACH --------------- #
+# Load Data
+power_usage <- read_excel("C:/Users/RANDUL/Desktop/2nd Semester of 2nd Year/5DATA001C.2 Machine Learning and Data Mining/CW/UoW_load.xlsx")
+View(power_usage)
+
+#09:00 , 10: 00, 11:00
+power_usage <- power_usage$`11:00`
+View(power_usage)
+str(power_usage)
+plot(power_usage)
+
+# Build the Partial Auto Correlation plot
+pacf(x = power_usage, plot = TRUE)
+
+# Create Lags for Time Series
+lag1 = lag(power_usage,1)
+lag2 = lag(power_usage,2)
+lag3 = lag(power_usage,3)
+lag4 = lag(power_usage,4)
+lag5 = lag(power_usage,5)
+power_usage <- cbind(power_usage,lag1,lag2,lag3,lag4,lag5)
+
+# Formatting power_usage
+power_usage <- na.omit(power_usage)
+sum(is.na(power_usage))
+
+colnames(power_usage) <- c("wineD_original_data","v2","v3","v4","v5","v6")
+
+# Scaling Data
+scaled_poweruse <- scale(power_usage)
+str(scaled_poweruse)
+
+# Save the processed data to excel file
+# For Excel files separately 09:00 , 10:00, 11:00 hours
+write.xlsx(scaled_poweruse,
+           file = "C:/Users/RANDUL/Desktop/2nd Semester of 2nd Year/5DATA001C.2 Machine Learning and Data Mining/CW/scaled_powerusage_09.xlsx",
+           col.names = TRUE, append = TRUE, row.names = FALSE)