File from a situation where it is required to predict today’s stockprice from the stock prices of the previous three days:
Could you use a decision tree classifier for this task? Why or why not?
2 Answers
$\begingroup$
$\endgroup$
As explained in here, decision trees have their disadvantages of a high chance of overfitting, instability with little variance, and low bias. Moreover, considering the fact that your data appears to be numerical, and continuous, there's a high chance that the decision tree might calculate the entropy of your features while only considering scenarios in your training data.
Decision trees tend to perform better in the scenarios where you have repetitive scenarios, and where your data is pretty much correlated, and most importantly, where you have discontinuous, categorical data in your features, which certainly isn't the case for the data which you mentioned.
$\begingroup$
$\endgroup$
1
LSTM, belonging to the family of RNN is best suited for this problem. Decision tree is not good in handling the sequencial data like stock compared to LSTM. LSTM network can be thought of memory networks which will store and process the previous data and pass it to till the last LSTM node to produce the output. In this case, LSTM can be used to store and process the information of past 3 days to predict the current stock price.
Infact Stock prediction is one of the most successful application of LSTM.
You can find a lot of resources in Google if you search for "Stock prediction using LSTM".
-
$\begingroup$ Do you have any evidence that an LSTM truly is the best algorithm for stock prediction? I'm not saying it isn't competitive, but there are many algorithms for modeling time series regressions and it's bold to definitively claim one algorithm's superiority. $\endgroup$– Alex LJan 2, 2019 at 21:05