Ames house price

#check with  lm cv=10
#Linear Regression 
#2346 samples
#  46 predictor
#Train:
#  RMSE      Rsquared   MAE     
#  29918.51  0.8569286  18859.15
library(AmesHousing)
data("ames_raw")
ames.ds=ames_raw
dim(ames.ds)
str(ames.ds)
ames.ds=data.frame(ames.ds)
 
#Identify missing values
sum(is.na(ames.ds))
miss.val.col=sapply(colnames(ames.ds),function(x) sum(is.na(ames.ds[,x]))) #by columns
barplot(miss.val.col[miss.val.col>nrow(ames.ds)/10],horiz=FALSE)
rm.col=which(miss.val.col>nrow(ames.ds)/2)  
missing_plot(ames.ds)
vis_miss(ames.ds,warn_large_data=FALSE)
#remove columns with more than 50% missing values
ames.ds=ames.ds[,-rm.col]
#missing value imputation using knn
ames.df=kNN(ames.ds,addRF = TRUE,variable = colnames(ames.ds),k=5)[,1:ncol(ames.ds)]
#feature selection using correlation 
ames.d=lapply(ames.df, function(x) {
  if(is.factor(x)) as.numeric(as.character(x)) else x
})
ames.d=lapply(ames.d, function(x) {
  if(is.character(x)) as.numeric(as.factor(x)) else x
})
ames.d=data.frame(ames.d)
T1=makeRegrTask(data =ames.d , target = "SalePrice")
feature_val = generateFilterValuesData(T1,method = "FSelectorRcpp_information.gain",equal=TRUE)
filtered.task = filterFeatures(T1, method = "FSelectorRcpp_information.gain", perc = 0.6,equal=TRUE)
filt_features=as.vector((getTaskFeatureNames(filtered.task)))
fnl_ds=ames.d[,c(filt_features,"SalePrice")]
#Splitting dataset
train_index=createDataPartition(fnl_ds$SalePrice,p=0.8,list=FALSE)

train_data=fnl_ds[train_index,filt_features]
train_price=fnl_ds$SalePrice[train_index]
test_data=fnl_ds[-train_index,filt_features]
test_price=fnl_ds$SalePrice[-train_index]
#Check distribution of data 

linear.model=train(train_data,train_price,
                   method="lm",
                   trControl = trainControl(method = "cv",number=10))
summary(linear.model)
plot(linear.model$results)                 
p.test=predict(linear.model,test_data)

rmse=sqrt(mean(p.test-test_price)^2)