TimeSeries prediction with LSTM Neural Network with (Keras and Pytorch)

S&P 500 Stock Price Prediction

1. Detailed analysis of Stock prices as time-series data to find insights and findout best stocks among all the stocks

2. Visualized stock prices and prove the hypothesis of daily return of stocks

3. Model evaluation metrics & performance comparison with RMSE & MAPE to measure accuracy of our model.

4. Analyize stocks historical data with Simple Moving average(SMA) & Exponential Moving average (EMA) techniques to predict stock movements.

   • Exponential Moving Average (EMA) is measuring trend direction over a period of time
   • SMA simply calculates an average of price data,

5. Apply ML algorithms & prepare stocks historical dataset for model training

6. LSTM (Long Short-Term Memory) Recurrent Neural network model of deep learning framework for time-series data for model training and predictions,

• When evaluating a model's accuracy using RMSE (Root Mean Square Error) and MAPE (Mean Absolute Percentage Error), you are essentially comparing how far, on average, your model's predictions deviate from the actual values,

• with RMSE giving more weight to larger errors due to its squared calculation,

• while MAPE expresses the error as a percentage of the actual value, making it useful for understanding relative error size across different scales;

• a lower value in both metrics indicates a more accurate model

  A lower RMSE indicates a better fit overall, while a lower MAPE means a smaller average percentage error.

7. Evaluating the performance of a regression model

• When evaluating the performance of a regression model, "accuracy" is not typically used as the primary metric, instead,

• Metrics like Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R-squared are used to assess how closely the model's predictions align with the actual values,

• As regression models predict continuous values rather than discrete classifications where "accuracy" is more applicable; essentially,

• You are measuring in regression how "close" the predictions are to the true values, not just whether they are "correct" or "incorrect" like in classification tasks.

• Key points about evaluating regression models - Regression metrics:

  1. Mean Squared Error (MSE): Calculates the average of the squared differences between predicted and actual values.

  2. Root Mean Squared Error (RMSE): The square root of MSE, providing a more interpretable error value in the original unit of the data.

  3. Mean Absolute Error (MAE): Calculates the average absolute difference between predicted and actual values, less sensitive to outliers than MSE.

  4. R-squared (coefficient of determination): Represents the proportion of variance in the dependent variable explained by the independent variables.

  Why use not accuracy in regression:

  • Continuous values: Regression models predict continuous values, not discrete classes, so a simple "correct/incorrect" classification doesn't apply.

  • Magnitude of error matters: In regression, the magnitude of the error is important, not just whether the prediction is "correct" or not