好的，根据您的需求，我为您编写了如下代码。
首先，我们需要加载所需的包和数据：
```R
library(xlsx)
library(randomForest)
library(keras)
library(e1071)
data_path <- "D:/Desktop/002362hwkj.xlsx"
stk_data <- read.xlsx(data_path, sheetIndex = 1)
```
然后，我们将数据按时间升序排列，并将日期列转化为日期格式：
```R
stk_data$Date <- as.Date(stk_data$Date)
stk_data <- stk_data[order(stk_data$Date),]
```
接下来，我们将数据划分为训练集和测试集。由于您没有指定测试集的时间范围，这里我将最后一年的数据作为测试集：
```R
train_data <- stk_data[1:(nrow(stk_data)-252),]
test_data <- stk_data[(nrow(stk_data)-251):nrow(stk_data),]
```
然后，我们为每个模型准备数据。对于随机森林和支持向量机，我们只需要选择需要的特征并将其转化为矩阵格式。对于长短期记忆神经网络，我们需要将数据转化为时间序列格式，并将其标准化。
```R
# 随机森林和支持向量机
train_rf_svm <- train_data[,c("Open", "Close", "Volume")]
train_rf_svm <- data.matrix(train_rf_svm)
test_rf_svm <- test_data[,c("Open", "Close", "Volume")]
test_rf_svm <- data.matrix(test_rf_svm)
# 长短期记忆神经网络
train_lstm <- train_data[,c("Close")]
train_lstm <- data.matrix(train_lstm)
scaler <- preProcess(train_lstm, method = c("scale", "center"))
train_lstm <- predict(scaler, train_lstm)
test_lstm <- test_data[,c("Close")]
test_lstm <- data.matrix(test_lstm)
test_lstm <- predict(scaler, test_lstm)
train_lstm <- array(train_lstm, dim = c(nrow(train_lstm), 1, 1))
test_lstm <- array(test_lstm, dim = c(nrow(test_lstm), 1, 1))
```
接下来，我们分别训练随机森林、长短期记忆神经网络和支持向量机，并对测试集进行预测：
```R
# 随机森林
rf_model <- randomForest(train_rf_svm[,1:2], train_rf_svm[,3], ntree = 500)
rf_pred <- predict(rf_model, test_rf_svm[,1:2])
# 长短期记忆神经网络
lstm_model <- keras::keras_model_sequential()
lstm_model %>%
layer_lstm(units = 50, input_shape = c(1, 1)) %>%
layer_dense(units = 1)
lstm_model %>% compile(loss = 'mean_squared_error', optimizer = 'adam')
lstm_model %>% fit(train_lstm[,1:1,], train_lstm[,1,], epochs = 100, batch_size = 32)
lstm_pred <- predict(lstm_model, test_lstm)[,1]
# 支持向量机
svm_model <- svm(train_rf_svm[,1:2], train_rf_svm[,3], kernel = "linear")
svm_pred <- predict(svm_model, test_rf_svm[,1:2])
```
最后，我们计算三个模型的准确率，并绘制时间序列图：
```R
# 计算准确率
rf_accuracy <- sum(rf_pred == test_rf_svm[,3])/nrow(test_rf_svm)
lstm_accuracy <- sum(round(lstm_pred) == test_lstm[,1])/nrow(test_lstm)
svm_accuracy <- sum(svm_pred == test_rf_svm[,3])/nrow(test_rf_svm)
cat("随机森林准确率：", rf_accuracy, "\n")
cat("LSTM准确率：", lstm_accuracy, "\n")
cat("支持向量机准确率：", svm_accuracy, "\n")
# 绘制时间序列图
par(mfrow=c(3,1))
plot(test_data$Date, test_data$Close, type = "l", col = "blue", xlab = "", ylab = "Price")
lines(test_data$Date, rf_pred, col = "red")
legend("bottomright", legend = c("Actual", "Predicted"), col = c("blue", "red"), lty = c(1, 1))
plot(test_data$Date, test_data$Close, type = "l", col = "blue", xlab = "", ylab = "Price")
lines(test_data$Date, round(lstm_pred), col = "red")
legend("bottomright", legend = c("Actual", "Predicted"), col = c("blue", "red"), lty = c(1, 1))
plot(test_data$Date, test_data$Close, type = "l", col = "blue", xlab = "", ylab = "Price")
lines(test_data$Date, svm_pred, col = "red")
legend("bottomright", legend = c("Actual", "Predicted"), col = c("blue", "red"), lty = c(1, 1))
```
注意：由于数据未提供证券代码，代码中的模型训练和预测均基于数据的 Close 列。如果您的数据中包含多个证券的交易数据，请修改代码以适配您的数据。
扫一扫
179
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
