Practical Neural Networks & Deep Learning In R

Apply neural networks to binary credit default classification using predictors like payment history, bill amount, age, education, and gender, with a 75/25 split and 10-fold cross-validation achieving about 0.818 accuracy.

Implement a multi-layer perceptron for supervised classification in R, from data preparation and train-test split to backpropagation training, prediction on unseen data, and evaluation with a confusion matrix.

Demonstrates using neural networks for multiclass classifications with iris data, applying softmax modeling, predicting species from predictors, and evaluating results with confusion matrices and accuracy.

Learn to build a multiclass neural network for handwritten digit images, loading pixel predictors, converting data to matrices, and predicting unseen digit labels (0–9) from training and test sets.

Explore neural network based regression to predict cement strength from predictors like slag, ash, water, cement, and age using the neuralnet package, with 0-1 normalization and a 75/25 train-test split.

Identify variable importance in neural networks using neural interpretation diagrams and the Gonzalez and OED algorithms, applied to concrete strength prediction with predictors such as cement, water, and aggregate.

Build a two-hidden-layer deep neural network in R with neuralnet for binary classification on college data. Normalize predictors, encode the target, and achieve 93% accuracy via a confusion matrix.

Implement a simple deep neural network in R using the deepnet package for regression on the Boston housing data, including train-test splitting, normalization, and mean squared error evaluation.

Learn to read external data in H2O for neural networks and deep learning in R, including starting H2O, configuring memory, importing files, and splitting data into training and testing sets.

Build deep neural networks in R with H2O for multi-class fashion item classification, using three hidden layers with dropout and cross-entropy loss to reach 91% accuracy on a 75/25 split.

Identify the most important predictors, such as radius mean and texture mean, for distinguishing malignant versus benign tumors using h2o deep learning and assess AUC performance.

Explore activation functions in neural networks, including sigmoid, softmax, and ReLU, their roles in non-linear modeling, probability outputs, and multiclass or binary classification in CNNs.

Explore regression with deep learning by implementing a two-hidden-layer H2O deep learning model on the Boston housing dataset, predicting prices from 13 predictors and evaluating error to refine architecture.

Implement an autoencoder deep learning model in R using h2o to detect credit card fraud, using predictors X1–X30 and 31st column as the label, an unsupervised approach with train-test split.

Explore unsupervised classification with autoencoders in R, using H2O deep features to distinguish binary and malignant diagnoses and visualize results with cube plots.

Install and use mxnet in R and RStudio to accelerate deep learning analyses on personal laptops. Learn installing via the base package and loading mxnet for workflows.

Implement a binary classification using mxnet's maxnet package, building a multi-layer perceptron with 60 predictors, training on 75% of data and evaluating with softmax activation.

Implement an ann-based regression using MXNet with the Boston housing dataset, building train and test matrices, a hidden layer, and linear regression output, evaluating with mean squared error.

Learn to build a deep neural network for multi-class fashion item classification with MXNet, using a 128–64–10 fully connected architecture, reglue and softmax activations, and a 75/25 data split.

Evaluate the accuracy of a deep neural network built with three hidden layers and a softmax output, using a confusion matrix and testing data to gauge generalizability.

Implement MXNet via caret to build a neural network (mlp) for classifying malignant versus benign tumors, including data prep, train-test split, mlp grid specification, and evaluation with a confusion matrix.

Explore convolutional neural networks (CNNs) for image classification, learning high-level features via convolution, nonlinearity, pooling, and classifying with fully connected layers and softmax.

Implement a convolutional neural network for multi-class image classification. Use 5x5 convs, relu activations, max pooling, and softmax for 10 classes, achieving ~89% accuracy on unseen fashion data.

Learn to process real images with MXNet by resizing to 28 by 28 grayscale, converting to 784-pixel vectors, labeling cats and dogs, and training a deep neural network for classification.

Explore recurrent neural networks for forecasting temporal data by predicting humidity from date-based predictors, training on historical data, and comparing predicted versus actual humidity to demonstrate convergence.

See how GitHub acts as your data portfolio, and use GitHub desktop to push code from your local folder, create repositories with readme, and collaborate with others.