Data Science Project

*Context
Irrespective of whether or not data and images are stored/analyzed in a centralized
manner, variability in scanner models, acquisition protocols and reconstruction
settings are unavoidable in the current clinical practice. Yet radiomics are notoriously
sensitive to such protocol variations. Hence, there is a clear need for the harmonization
of features in order to allow consistent findings in radiomics multicenter studies.

*Objective
The objective of this project is to develop different models to predict failure (endpoint)
of the radiomics signature based from MRI, PET and CT scans.

*Dataset
radiomics.csv contains 197 rows and 498 columns:
Failure.binary: binary property to predict
You can split the dataset as you want to create the training/validation/test datasets

*Models
You have to deliver three different models:
-Model1
§ Create an ensemble classification model (atleast 3 models of your choice).
§ Preprocess the data
o Check for null and missing values
o Check for normality, if not, normalized the data
o Get the correlation of the whole data expect the categorical variables
§ Split the data into training (80%) and testing (20%)
§ Print the AUC values during Training
§ Print the Top 20 important features during Training
§ Print the AUC values during Testing

-Model2
§ Create a neural network-based classification model.
§ Create five hidden layers with 256, 128, 128, 64 and 64 neurons, respectively
with activation functions of Sigmoid

§ Create an output layer with ten neurons respectively with activation functions
of Softmax.
§ Every layer is followed by a dropout to avoid overfitting.
§ Copy the slide 15 backpropagation compiler approach.
§ Copy the slide 33 model compiler approach.
§ Train the model with epoch = 10, batch size = 128 and validation split = 0.15
(reference slide 33).
§ Evaluate the trained model using the testing dataset.
§ Get the model prediction using the testing dataset.

-Model3
§ Without considering the binary output and categorical variables in the dataset,
compare the following clustering technique results:
o K-Means
o Hierarchical
o Model Based

*Application
To deliver:
§ 3 (Model1, Model2, and Model13) R Markdown
§ 3 (Model1, Model2, and Model13) PDF Files from an R Markdown Outputs
§ These should be pushed into your final github repository.
§ Name the repo as INFS692

Readme file
This md file must have the documentation about your application, models, packaging,
setup and dockerization

Git
Use git to version your code and push it to any public repository and send/upload the
link in the My Drive before December 10, 2022 at 11:59PM.

Evaluation
You will be evaluated on:
§ Quality and structure of your codes
§ Models architecture
§ Git commits quality
§ Data preparation and preprocessing
§ Documentation
§ The quality of your answers