if loading fail on the github, please go to this link to see the code online: https://nbviewer.jupyter.org/github/Bucky-Cheng/Stock-Prediction-LSTM-AAPL/blob/master/LSTM_Stock_AAPL_G_1.ipynb
1.import data
2.data cleaning
3.add technical indicators
4.technical indicators visualization
5.technical indicators interpretation
6.more feature engineering
7.rebuild dataset, split train/val/test set
8.create model(LSTM)
9.train model, prediction
10.compare prediction
11.conclusion and future
Date
Open
High
Close
Adj Close
Day of week
Close price mean
Close price Variance
Close price tandard deviation
Stochastic Relative Strength Index (StoRSI) and RSI
%K and %D
Chaikin Money Flow(CMF)
Parabolic SAR
Rate of Change
Volume Weighted Average Price (VWAP)
Momentum 10Days
Moving Average Convergence Divergence(MACD)
Average Directional Movement Index (ADX)
Williams %R
The Ichimoku Cloud
        Conversion Line
        Base Line
        Leading Span A
        Leading Span B
        Cloud(Leading Span A, Leading Span B)
Bollinger Bands
Date
Open
High
Close
Adj Close
Data 1 MSE:0.0005706506187007371
Data 2 MSE:0.0008784179054824572
The line chart illustrate the difference clearly. The data1 is much better than data2, and data2 line is like a stright line after 5th, which have not any help, the price will remain same as previous day. So, the technical indiacators are prvide the meaningful features for the price predication.
But, there are some problems, the trend can't remain the one direction 2days, like the 13th-15th, when up to 14th from 13th, the trend change the direction to bottom, but the truth remian upward. And the 20th, the price has a big rising, but model can not to predict this big rising, and this the limitation of the dataset which only have price.
The model can not to predict big suddenly increasing, so I think we need to use the News or Filings to identify this situation, the 20th is 30/07/2020, and Apple reported Third Quarter Results Reports, which is called a historically strong quarter, a lot of News report this and all have the positive sentiments, for example, the News of CNBC which said "Apple posts blowout third quarter, with sales up 11% despite coronavirus disruptions "(URL: https://www.cnbc.com/2020/07/30/apple-aapl-earnings-q3-2020.html).
So, I believe the News, filings and reports are significant features for stock predication, and I will using NLP to research it ,for example use BERT to analysis sentiments.
Data 1 Close Price MSE:90.61528315553844
Data 2 Close Price MSE:135.424431323489