Both social media sentiment and stock market data are crucial for stock price prediction. So, in this project we analyzed the dynamics of stock markets based on both social media news (text data) and stock prices (numerical data).
The dataset we are working on is a combination of Wallstreetbets-Reddit news and the Standard & Poor’s 500 (S&p 500) stock price from 2013 to 2018.
-
The news dataset contains the top 25 news from Reddit on each day from 2013 to 2018.
-
The S&P 500 contains the core stock market information for each day such as Open, Close, and Volume.
-
The SCORE of the dataset is whether the stock price is increase (labeled as 1) or decrease (labeled as 0) on that day.
Introduction:
- data dataset comprises 5698 rows and 8 columns.
- Dataset consists of continuous variable and float data type.
- Dataset column variables 'Open', 'Close', 'High', 'Low', 'Volume', are the stock variables from historical dataset and other variables are showing polarity of news which are the derived variables using sentiment analysis as discussed in the above section.
Descriptive Statistics:
Using describe() we could get the following result for the numerical features
open high low close volume count 5697.000000 5697.000000 5697.000000 5698.000000 5.698000e+03 mean 88.139399 89.012936 87.245609 88.146015 1.718703e+06 std 32.666995 32.960833 32.363413 32.660301 1.248357e+06 min 30.380000 31.090000 29.730000 29.940000 1.000000e+02 25% 64.650000 65.310000 64.053300 64.672500 9.880475e+05 50% 80.750000 81.490000 79.990000 80.750000 1.460298e+06 75% 105.270000 106.270000 104.350000 105.345000 2.135991e+06 max 201.240000 201.240000 198.160000 200.380000 3.378024e+07
We filled out the NaN values in the missed three topics. And got the polarity and subjectivity for the news' topics. Polarity is of 'float' type and lies in the range of -1, 1, where 1
means a high positive sentiment, and -1 means a high negative sentiment.
So, they will be very helpful in determining the increase or decrease of the stock market.
Then we checked the missing values in the stock market information, it was complete. Then we merged the sentiment information (polarity ) by date with the stock market information (Open, High, Low, Close, Volume, Adj Close) in merged_data dataframe.
Before modeling and after splitting we scaled the data using standardization to shift the distribution to have a mean of zero and a standard deviation of one.
from sklearn.preprocessing import StandardScaler
scaler = StandardScaler().fit(X_train)
rescaledX = scaler.transform(X_train)
rescaledValidationX = scaler.transform(X_valid)
fit_transform() is used on the training data so that we can scale the training data and also learn the scaling parameters of that data. Here, the model built by us will learn the mean and variance of the features of the training set. These learned parameters are then used to scale our test data.
transform() uses the same mean and variance as it is calculated from our training data to transform our test data. Thus, the parameters learned by our model using the training data will help us to transform our test data. As we do not want to be biased with our model, but we want our test data to be completely new and a surprise set for our model.
Now, after observing the outliers in polarity of a lot of topics, we decided to concatenate all the 14 topics in one paragraph, then we can get only one column for polarity. So, we merged these data again with the stock market numerical information and got merged_data dataframe, then scaled it.
Accuracy , Precision , Recall and F1 Score
- Accuracy: What proportion of actual positives and negatives is correctly classified?
- Precision: What proportion of predicted positives are truly positive ?
- Recall: What proportion of actual positives is correctly classified ?
- F1 Score : Harmonic mean of Precision and Recall
- Logistic Regression helps find how probabilities are changed with actions.
- The function is defined as P(y) = 1 / 1+e^-(A+Bx)
- Logistic regression involves finding the best fit S-curve where A is the intercept and B is the regression coefficient. The output of logistic regression is a probability score.
After choosing model based on confusion matrix here where choose the features taking in consideration the deployment phase.
We know from the EDA that all the features are highly correlated and almost follow the same trend among the time. So, along with polarity and subjectivity we choose the open price with the assumption that the user knows the open price but not the close price and wants to figure out if the stock price will increase or decrease.
When we apply the logistic regression model accuracy dropped from 80% to 55%. So, we will use both Open and Close and exclude High, Low, Volume, Adj Close.
precision recall f1-score support
0 1.00 1.00 1.00 2563950
1 0.00 0.00 0.00 968
accuracy 1.00 2564918
macro avg 0.50 0.50 0.50 2564918
weighted avg 1.00 1.00 1.00 2564918