Netflix Stock Price Prediction:
1. Import Libraries
from flask import Flask, render_template, request, send_from_directory
import os
import datetime
import pandas as pd
import numpy as np
import joblib
from tensorflow.keras.models import load_model
import matplotlib.pyplot as plt
import matplotlib.dates as mdates
• Flask: Manages routes, handles user input, and serves templates.
• os: Manages file directories for saving prediction images.
• datetime: Handles date and time operations.
• pandas: Processes stock data.
• numpy: Handles numerical data for predictions.
• joblib: Loads the scaler used for data normalization.
• load_model: Loads the pre-trained LSTM model.
• matplotlib: Generates prediction graphs.
• mdates: Formats date labels in the prediction graph.
2. Initialize Flask App and Configurations
app = Flask( name )
app.config['PREDICTIONS_FOLDER'] = 'static/predictions'
• Creates a Flask application instance.
• Sets a folder to store prediction graphs.
3. Load Model and Data
# Load model and scaler
model = load_model('lstm_model.h5')
scaler = joblib.load('scaler.pkl')
# Load and preprocess the dataset
data = pd.read_csv('NFLX_dataset.csv')
data['Date'] = pd.to_datetime(data['Date'])
data.set_index('Date', inplace=True)
data = data[['Close']]
• Loads the LSTM model (lstm_model.h5) and the scaler (scaler.pkl).
• Reads the stock dataset (NFLX_dataset.csv), converts the Date column to
datetime, and sets it as the index.
• Keeps only the Close column (closing prices).
4. Feature Engineering
# Feature Engineering (same as during training)
data['5_MA'] = data['Close'].rolling(window=5).mean()
data['30_MA'] = data['Close'].rolling(window=30).mean()
data['Volatility'] = data['Close'].rolling(window=5).std()
data['Returns'] = data['Close'].pct_change()
data.dropna(inplace=True)
• Calculates additional features required for predictions:
– 5_MA: 5-day moving average.
– 30_MA: 30-day moving average.
– Volatility: 5-day standard deviation of prices.
– Returns: Percentage change in price.
• Removes rows with missing values introduced by moving averages.
5. Scale Data
# Scale the data with all features
data_scaled = scaler.transform(data[['Close', '5_MA', '30_MA',
'Volatility', 'Returns']])
• Uses the preloaded scaler to normalize the dataset.
6. Define Custom Prediction Function
def predict_custom_dates(custom_dates):
custom_dates_datetime = []
for date in custom_dates:
try:
custom_dates_datetime.append(datetime.datetime.strptime(date, '%d-%m-
%Y'))
except ValueError:
print(f"Invalid date format: {date}. Skipping this date.")
if not custom_dates_datetime:
return [], []
# Prepare input for prediction
last_lookback_data = data_scaled[-120:]
last_input = np.array([last_lookback_data])
custom_predictions = []
for _ in custom_dates_datetime:
prediction = model.predict(last_input)
custom_predictions.append(prediction[0, 0])
prediction_padded = np.concatenate((prediction, np.zeros((1,
4))), axis=1)
prediction_reshaped = prediction_padded.reshape(1, 1, 5)
last_input = np.concatenate([last_input[:, 1:, :],
prediction_reshaped], axis=1)
custom_predictions = np.array(custom_predictions).reshape(-1, 1)
custom_predictions =
scaler.inverse_transform(np.concatenate((custom_predictions,
np.zeros((custom_predictions.shape[0], 4))), axis=1))[:, 0]
return custom_dates_datetime, custom_predictions
• Parses user-provided dates.
• Prepares the last 120 scaled data points for predictions.
• Iteratively predicts prices for each date using:
– Current model input.
– Updated predictions for the next step.
• Converts predictions back to the original scale.
7. Generate Prediction Image
def save_prediction_image(predicted_dates, predicted_prices):
if not os.path.exists(app.config['PREDICTIONS_FOLDER']):
os.makedirs(app.config['PREDICTIONS_FOLDER'])
timestamp = datetime.datetime.now().strftime("%d%m%Y_%H%M%S")
prediction_filename = f"prediction_{timestamp}.png"
filepath = os.path.join(app.config['PREDICTIONS_FOLDER'],
prediction_filename)
# Plot actual data and predictions
plt.figure(figsize=(14, 8))
last_month_data = data[-120:]
last_month_actual = scaler.inverse_transform(data_scaled[-
120:, :])[:, 0]
plt.plot(last_month_data.index, last_month_actual, label='Actual
Price (Last 120 Days)')
plt.plot(predicted_dates, predicted_prices, label='Forecasted
Prices', marker='o')
plt.legend()
plt.savefig(filepath)
plt.close()
return prediction_filename
• Creates a graph showing:
– Actual prices for the last 120 days.
– Predicted prices for custom dates.
• Saves the graph in the predictions
folder.
8. Route for Home Page
@app.route('/', methods=['GET', 'POST'])
def index():
last_120_days_data = data[-120:]
first_price = last_120_days_data['Close'].iloc[0]
last_price = last_120_days_data['Close'].iloc[-1]
price_change = last_price - first_price
percentage_change = (price_change / first_price) * 100
insights = {
'first_price': round(first_price, 2),
'last_price': round(last_price, 2),
'price_change': round(price_change, 2),
'percentage_change': round(percentage_change, 2),
'trend': "rose" if price_change > 0 else "fell"
}
if request.method == 'POST':
date_input = request.form.get('dates')
custom_dates = [date.strip() for date in
date_input.split(",")]
predicted_dates, predicted_prices =
predict_custom_dates(custom_dates)
formatted_dates = [date.strftime('%d-%m-%Y') for date in
predicted_dates]
predictions = list(zip(formatted_dates, predicted_prices))
image_url = save_prediction_image(predicted_dates,
predicted_prices)
return render_template('index.html', image_url=image_url,
predictions=predictions, insights=insights)
return render_template('index.html', image_url=None,
insights=insights)
• GET Request: Displays the form and insights.
• POST Request:
– Processes input dates.
– Calls prediction and graph-generation functions.
– Passes results to the index.html template.
9. Serve Images
@app.route('/static/predictions/<filename>')
def display_image(filename):
return send_from_directory(app.config['PREDICTIONS_FOLDER'],
filename)
• Serves prediction images dynamically.
10. Run the App
if name == ' main
':
app.run(debug=True)
• Starts the app in debug mode for easier development.
