You are on page 1of 12

Enhancing Stock Market

Forecasting with LSTM-based Deep


Learning Models.

❏ TEAM MEMBERS
❏ PRANAV SARAVANAN - 20BCE0974
❏ KEERTHEESWAR - 20BDS0304 CAPSTONE GUIDE:
❏ SYED SHA SUHEB - 20BDS0394 DR.SAURABH AGRAWAL
Overview
● We are analyzing the predictive power of LSTM networks for Forex and stock
market movements, concentrating just on LSTM for a thorough examination.
● We accomplish this by collecting and carefully cleaning historical data, creating
customized LSTM models for forecasting, assessing performance indicators, and
analyzing the outcomes to extract useful information.
● Our results demonstrate that LSTM is a powerful instrument for precisely
predicting dynamic market movements, which may be used to optimize decision-
making processes in financial institutions.
Literature Review
Deep Learning for Stock Market Prediction

● Deep Learning (DL) techniques are increasingly used for stock market
prediction.
● DL models like CNNs, LSTMs, and TCNs show promise in capturing
complex market patterns.
● Data limitations, market volatility, model interpretability.
● Combining DL with other data sources (sentiment analysis), enhancing
interpretability for better decision-making.

Link for Literature Review:


Literature Review
Gaps

● Limited Integration of Data Modalities: Existing models struggle to


effectively integrate multiple data sources such as sentiment, price, and
attention measures.
● Lack of Interpretability: Many models lack clear explanations for
predictions, hindering investors' understanding of underlying assumptions.
● Inadequate Feature Extraction: Traditional models rely on customized
features, potentially missing underlying patterns in stock market data.
The Solution

● Effective Integration of Data Modalities: LSTM networks seamlessly integrate


sentiment, price, and attention measures, boosting predictive capabilities.
● Enhanced Interpretability: The model gains clarity, offering insights into
predictions and overcoming challenges in financial forecasting understanding.
● Efficient Handling of Sequential Data: LSTM efficiently learns patterns, adapts
to market changes without extensive preprocessing, yielding accurate predictions
and capturing market dynamics.
Introduction

Our choice of LSTM neural networks stems from their proficiency in handling
sequential data and capturing temporal dependencies, making them ideal for financial
time series forecasting. Leveraging LSTM's Long-Term Memory capability enhances
accuracy in stock price forecasts by adapting to shifting market conditions and
grasping long-term trends. Moreover, by incorporating advanced techniques like
wavelet transform and attention mechanisms, we aim to harness the power of deep
learning to provide investors with more accurate predictions and valuable insights in
volatile financial markets.
Methodology Adapted

In our methodology, we've chosen LSTM networks for stock market


prediction due to their proven ability to capture temporal dependencies in
sequential data. Unlike other methods, LSTM's design allows it to handle
sequential data seamlessly, adapting to market conditions without extensive
preprocessing. With its long-term memory capacity, LSTM enhances accuracy
by effectively capturing the complex dynamics of financial markets.
STEPS OF THE WORK
● Data Input: Load historical stock prices from a CSV file
named "chinese-stock-dataset.csv" for model training.
● Data Preprocessing: Convert data into numpy arrays, create
sequences of data points, split into training and testing sets,
normalize using a standard scaler, and reshape for LSTM input.
● Model Building: Define LSTM model architecture using the
Keras Sequential API, incorporating LSTM layers, dropout
layers, and a dense output layer.
● Model Training: Train the model using the preprocessed
training data, specifying batch size, epochs, validation split,
and verbosity level.
● Model Evaluation: Evaluate the trained model's performance
on preprocessed testing data by comparing its predictions to
actual values.
IMPLEMENTATION (ABOUT THE DATASET)

Description of the Dataset

This dataset contains daily stock market data including open, high, low,
volume, amount, and close prices. The data covers the price movements and
trading volumes of a specific stock over a period of time, making it suitable
for training LSTM models for stock price prediction tasks.

DATASET LINK: dataset


IMPLEMENTATION

Collab link : Code


RESULT

Overall accuracy: 80.09%(Test) , 91.84%(Train)

LSTM Model Accuracy Train Score Test Score


Metrics

Mean Squared Error (MSE) 0.08161179842342596 0.27907556614879053

Root Mean Squared Error (RMSE) 0.285677787766963 0.5282760321543942

Mean Absolute Error (MAE) 0.17383319195430777 0.39010961703281677


Thank You…

You might also like