Background

Cirrhosis results from prolonged liver damage, leading to extensive scarring, often due to conditions like hepatitis or chronic alcohol consumption. The data provided is a subset sourced from a Mayo Clinic study on primary biliary cirrhosis $($PBC$)$ of the liver carried out from $1974$ to $1984.$

This is a dataset to develop and validate machine learning algorithms for predicting the survival status of the collected patients. There are $312$ patients in the data set $(224$ for train and $88$ for test$),$ and each patient has $17$ collected features. The aim of this task is to utilize $17$ clinical features for predicting survival state of patients with liver cirrhosis. The survival states include $0=$ D $($death$),1=$ C $($censored$),2=$ CL $($censored due to liver transplantation$)$

Specifically, the problem you are going to solve is: Can you

$$

Accurately predict the survival status given the labelled data?

$$

Well explain your prediction and the associated findings? For example, identify the key factors which are strongly associated with the response variable, i$.$e$.,$ survival status.

Data set

The training data contains $224$ rows and the test data contains $88$ rows, each of which have $19$ columns $($excluding the ID column$)$: the N$\_$Days attribute is the number of days between registration and the earlier of death, transplantation, or study analysis time in July $1986,$ the status attribute is the target variable that we will predict, and the rest $17$ columns can be used as the input features. The details of the original data set can be found and downloaded in the original UCI repository. The values of the $$status$$ column in the test set is leaved with empty to simulate real world predictions.

Evidence of Learning:

Execute your code into a jupyter notebook $(.$ipynb file$)$ and keep the output, write a report $(.$pdf file$)$ to answer the following questions, and submit your code and report to OnTrack.

$1.$

Load and explore the training and test dataset, do necessary pre$-$processing.

a$.$

Show both training and test dataset size.

b$.$

Based on the training and test data, show the feature types, and indicate which feature has missing values.

c$.$

Use an appropriate method to deal with the missing values for both the training and test set.

d$.$

Do necessary encoding for the categorical features.

e$.$

Show the label distribution based on the training data, is it a balanced training set?

$2.$

Based on the pre$-$processed training data from question $1,$ create three supervised machine learning $($ML$)$ models for predicting $$Status$.$

a$.$

Use an appropriate validation method, report performance score using a suitable metric. Is it possible that the presented result is an underfitted or overfitted one? Justify.

b$.$

Justify different design decisions for each ML model used to answer this question.

c$.$

Have you optimised any hyper$-$parameters for each ML model? What are they? Why have you done that? Explain.

d$.$

What can you do with the label imbalance issue?

e$.$

Finally, make a model recommendation based on the reported results and justify it

Use the best model that you get from question $2,$ do prediction on the pre$-$processed test set. Save your prediction $($the prediction should contain two columns only: testID and Status$),$ and submit it to the specific Kaggle in$-$class platform, do a screenshot of your model performance and report it$.$

Please answer all of question $1,$ I'll then use it for question $2$ and $3$