Materials from Neural Networks for Machine Learning Applications course as a part of Metropolia University of Applied Sciences Health Technology field.
Case 0. Machine Learning Basics
Creating a neural network model for hand-written digit classification. Getting familiar with the tensorflow keras library, Jupyter Notebook environment as well as code-docummenation best practises.
Case 1. Heart Disease Classification
Creating a sequential neural network model for the heart disease classification problem. Solution based on the 253,680 medical survey-responses dataset collected from The Behavioral Risk Factor Surveillance System 2015. Insightful data preprocessing, modeling and training a sequential neural network structure, finding the most efficient parameters for the network model. Performance evaluation and limiting the model overfitting.
Case 2. Pneumonia X-ray image analysis
Creating a convolutional neural network model for Pneumonia classification based on X-ray images classified as normal, bacterial or pneumonia. Image data preprocessing. Transfer learning: using predefined image-classification models such as MobileNetV2 and DenseNet169. Building convolutional models consisting of conv2D and pooling layers. Convolutional model training and performance evaluation focusing on sensitivity and specificity.
Building a recurrent neural network for text-reviews classification. Classifying text-reviews into three categories: negative, neutral and positive. Text-data preprocessing. Tokenization, sequentialization, stemming and lemmatization. Natural language processing. Modeling and training recurrent LSTM and convolutional neural networks. Evaluation and results comparison.
https://www.kaggle.com/czaacza/code
Private Metropolia Google Account