Summary of Deep Learning Models and Techniques in Projects

1. Linear Regression and Generalized Linear Model (GLM):

• Techniques: Logistic regression and linear regression models for predicting credit card default and startup success.
• Application: Used for binary classification (credit default) and general prediction tasks in startups.
• Model: Logistic regression with significant predictors, and linear regression for continuous data prediction (log volume).

2. Deep Neural Networks (DNN):

• Techniques: Feedforward neural networks with dropout regularization for credit card default classification.
• Application: Used to predict whether a customer will default.
• Model: A neural network with a single hidden layer and sigmoid activation for binary classification.

3. Convolutional Neural Networks (CNN):

• Techniques: Convolutional layers, max pooling, dropout, and fully connected layers for image classification.
• Application: Image classification on the Fashion MNIST dataset.
• Model: CNN architecture with two convolutional layers, followed by dense layers for multi-class classification.

4. Recurrent Neural Networks (RNN):

• Techniques: Simple RNN layers for sequential data, designed to capture temporal patterns.
• Application: Predicting time series data (log volume) from the NYSE dataset.
• Model: RNN with a single recurrent layer followed by a dense layer.

5. Long Short-Term Memory (LSTM):

• Techniques: LSTM units, a special type of RNN for learning long-term dependencies in sequential data.
• Application: Time series forecasting for air pollution levels in a multivariate setting.
• Model: LSTM network with multiple lag features for forecasting.