CODEWITHSUNDEEP
rmachine-learningsvmpredictionsentiment-analysis
james_ricky
james_ricky

Reputation: 17

Stock prediction + news sentiment with SVM in R?

I would like to predict the stock prices and news sentiment score together with SVM in R, in order to see whether news have an impact on stock price and their prediction. I read that support vector machines (svm) are a good machine learning approach for this problem. I have one column that represents the date of the stock and news, one column represents the stock prices on that day and 4 columns which represent the sentiment scores based on different lexica. I would like to test first with one of that lexica and if the models works, trying on the other. The dataset is included below. I found some examples with python but couldn't found something for R. I like to use the svm() function from the e1071 package

I split the data into train and test set:

sample <- sample(nrow(sentGI),nrow(sentGI)*0.70)
df.trainGI = sentGI[sample,]
df.testGI = sentGI[-sample,]

And I tried already this SVM code, but my wrong prediction rate is 100

plot(df.trainGI$GSPC.Close, df.trainGI$SentimentGI, pch = 19, col = c("red", "blue"))


svm_model_GI <- svm(SentimentGI.Class ~ ., df.trainGI)
print(svm_model_GI)

plot(svm_model_GI, df.trainGI)


svm_pred_GI <- predict(svm_model_GI, newdata = df.testGI, type="response")
rmse <- sqrt(mean((svm_pred_GI - df.testGI$GSPC.Close)^2))
rmse

What I am doing wrong here? Hope somebody can help me!

Dataset

Upvotes: 1

Views: 494

Answers (1)

wisamb
wisamb

Reputation: 502

You're using model accuracy to evaluate the model. Accuracy is used for classification problems but your response variable is continuous. You should use RMSE.

pred <- predict(radial.svm, newdata=df.test, type='response')
rmse <- sqrt(mean((pred - df.test$GSPC.Close)^2))
rmse

Continuation from comments:

enter image description here

The first plots GSPC.Close against date (left) and the second plots SentimentGI against date (right). Notice that stock prices generally increase over time whereas sentiment has a slope of 0 in that same time frame. What does that tell you?

Upvotes: 1

Related Questions