In [2]:
options(warn=-1)
library(caret)
library(openxlsx)
library(vegan)
library(randomForest)
library(class)
library(cluster)
library(e1071)
library(klaR)
library(rpart)
library(kknn)
#rm(list=ls(all=TRUE))
#ЗАГРУЗКА ИЗ EXSEL ПЕРВИЧНЫХ ДАННЫХ
N=1 # номер листа с данными
x <-read.xlsx("Region_Data4.xlsx", sheet = N)
#x
#ФОРМИРОВАНИЕ data.frame ОСНОВНЫЕ СОЦИАЛЬНО-ЭКОНОМИЧЕСКИЕ ПОКАЗАТЕЛИ РЕГИОНОВ РФ
xreg <-data.frame(P1=x$Pok_001,P2=x$Pok_002,P3=x$Pok_003,P4=x$Pok_004,
P5=x$Pok_005,P6=x$Pok_006,P7=x$Pok_007,P8=x$Pok_008,P9=x$Pok_009,P10=x$Pok_010,
P11=x$Pok_011,P12=x$Pok_012,P13=x$Pok_013,P14=x$Pok_014,P15=x$Pok_015,P16=x$Pok_016,
P17=x$Pok_017,P18=x$Pok_018)
#xreg
#Формирование фактора тип экономики региона
(xreg_f <- as.factor(x$Pok_018))
reg_ru = xreg[, -18]
y=xreg_f
#модель случайные деревья
train.index <- createDataPartition(y, p = .8, times = 20)
trControl = trainControl(method="oob", index = train.index)
print("Модель на основе исходного набора из 17 предикторов")
(modSource <- train(reg_ru, y, "rf", trControl = trControl))
#показатели важности переменных могут быть рассчитаны функцией varImp()
VagRf <-varImp(modSource, scale = FALSE)
x11()
plot(VagRf)
#Деревья на основе рекурсивного деления (recursive partitioning)
y=xreg_f
train.index <- createDataPartition(y, p = .8, times = 20)
trControl = trainControl(method="LGOCV", index = train.index)
print("Модель на основе исходного набора из 17 предикторов")
(modSource <- train(reg_ru, y, "rpart", trControl = trControl))
VagRp <-varImp(modSource, scale = FALSE)
x11()
plot(VagRp)
#Наивный байесовский классификатор
y=xreg_f
train.index <- createDataPartition(y, p = .8, times = 20)
trControl = trainControl(method="LGOCV", index = train.index)
print("Модель на основе исходного набора из 17 предикторов")
(modSource <- train(reg_ru, y, "nb", trControl = trControl))
VagNbc <-varImp(modSource, scale = FALSE)
x11()
plot(VagNbc)
#Алгоритм knn
w=c(10,11,14,16)
y=xreg_f
xx <-reg_ru
#xx <-xreg[,w]
train.index <- createDataPartition(y, p = .8, times = 20)
trControl = trainControl(method="LGOCV", index = train.index)
print("Модель на основе набора исходного набора из 17 предикторов")
(modSource <- train(xx, y, "kknn", trControl = trControl))
VagKNN <-varImp(modSource, scale = FALSE)
x11()
plot(VagKNN)
#Полученные результаты
- A
- T
- P
- A
- T
- P
- P
- A
- P
- T
- A
- P
- T
- A
- P
- P
- P
- D
- D
- D
- P
- P
- P
- D
- P
- A
- T
- A
- T
- T
- D
- P
- T
- T
- A
- T
- A
- T
- T
- A
- P
- P
- A
- P
- D
- P
- P
- T
- P
- D
- A
- D
- A
- P
- A
- P
- D
- P
- T
- D
- D
- A
- D
- D
- D
- D
- P
- D
- T
- D
- D
- D
- T
- D
- T
- D
- D
- D
- D
- D
- P
Levels:
- 'A'
- 'D'
- 'P'
- 'T'
[1] "Модель на основе исходного набора из 17 предикторов"
Random Forest 81 samples 17 predictors 4 classes: 'A', 'D', 'P', 'T' No pre-processing Resampling: Out of Bag Resampling Summary of sample sizes: 66, 66, 66, 66, 66, 66, ... Resampling results across tuning parameters: mtry Accuracy Kappa 2 0.5432099 0.3760150 9 0.6296296 0.4973107 17 0.5802469 0.4286307 Accuracy was used to select the optimal model using the largest value. The final value used for the model was mtry = 9.
[1] "Модель на основе исходного набора из 17 предикторов"
CART 81 samples 17 predictors 4 classes: 'A', 'D', 'P', 'T' No pre-processing Resampling: Repeated Train/Test Splits Estimated (20 reps, 75%) Summary of sample sizes: 66, 66, 66, 66, 66, 66, ... Resampling results across tuning parameters: cp Accuracy Kappa 0.05357143 0.5166667 0.34018849 0.17857143 0.4700000 0.25573559 0.28571429 0.3700000 0.08618519 Accuracy was used to select the optimal model using the largest value. The final value used for the model was cp = 0.05357143.
[1] "Модель на основе исходного набора из 17 предикторов"
Naive Bayes 81 samples 17 predictors 4 classes: 'A', 'D', 'P', 'T' No pre-processing Resampling: Repeated Train/Test Splits Estimated (20 reps, 75%) Summary of sample sizes: 66, 66, 66, 66, 66, 66, ... Resampling results across tuning parameters: usekernel Accuracy Kappa FALSE 0.4933333 0.3302650 TRUE 0.4633333 0.2856843 Tuning parameter 'fL' was held constant at a value of 0 Tuning parameter 'adjust' was held constant at a value of 1 Accuracy was used to select the optimal model using the largest value. The final values used for the model were fL = 0, usekernel = FALSE and adjust = 1.
[1] "Модель на основе набора исходного набора из 17 предикторов"
k-Nearest Neighbors 81 samples 17 predictors 4 classes: 'A', 'D', 'P', 'T' No pre-processing Resampling: Repeated Train/Test Splits Estimated (20 reps, 75%) Summary of sample sizes: 66, 66, 66, 66, 66, 66, ... Resampling results across tuning parameters: kmax Accuracy Kappa 5 0.58 0.4344736 7 0.58 0.4344736 9 0.57 0.4220641 Tuning parameter 'distance' was held constant at a value of 2 Tuning parameter 'kernel' was held constant at a value of optimal Accuracy was used to select the optimal model using the largest value. The final values used for the model were kmax = 7, distance = 2 and kernel = optimal.
