DataScience

This repo contains the Submission for the Data Science Module of the MSc Medical Informatics / Business Information Systems of Sarah Rebecca Meyer.

Assignment 1: Regression - Interest Rate Prediction

Objective

The task is to build a regression model for a peer-to-peer credit marketplace that predicts the appropriate interest rate for loan applicants. The model will be used to assist private lenders in determining a suitable rate based on the applicant’s available information.

Dataset

• The dataset is a modified version of the Lending Club Loan Dataset, which contains anonymized historical loan application data.
• Some attributes in the dataset are not available for new applications, so they must be excluded from training.
• Data preprocessing involves handling missing values, removing unnecessary attributes, and scaling numerical features.

Tasks

• Business Understanding & Data Exploration
  • Research financial lending concepts such as credit scores, debt-to-income ratios (DTI), and interest rates.
  • Explore the dataset to understand feature semantics, distributions, and correlations.
• Data Preprocessing:
  • Remove features unavailable in new applications (e.g., loan status, past payments).
  • Handle missing values and inconsistencies.
  • Normalize or standardize numerical variables if necessary.
• Model Training & Evaluation:
  • Train multiple regression models (e.g., linear regression, decision trees, ensemble methods).
  • Use Mean Squared Error (MSE) as the primary evaluation metric.
  • Apply cross-validation to select the best-performing model.
• Reality Check & Final Submission
  • The best model will be tested on a hidden dataset (secret data).
  • A separate R script is provided for the final evaluation.
  • The final model, training dataset, and all experimental scripts must be included in the submission.

Files

• final_main.Rmd: The final file where I performed the data processing and created the model.
• model_log.txt: In this file, I logged models upon creation. In this file, one gets a glimpse of the work I did and how I progressed / what I changed.
• reality_check.R: The file for the Reality Check. Upon execution, you will be prompted to select the dataset to run it.
• final_model_xgb.rds: The final XGBoost model with the best performance.

Assignment 2: Classification – Credit Card Customer Behavior Prediction

Objective

This assignment focuses on training a neural network classifier to predict how well a customer will repay their credit card debt based on application data. The goal is to improve risk assessment in credit lending.

Dataset

• The dataset is a modified version of the Credit Card Approval Dataset, containing anonymized customer applications and their repayment behavior.
• Features include demographics, financial status, and previous credit behavior.
• The target variable represents different delinquency levels (e.g., 30-59 days past due, 90-119 days overdue, etc.).

Tasks

• Data Exploration & Preprocessing
  • Convert categorical variables into numerical representations.
  • Normalize all input features to be within a [0,1] range, as required for neural networks.
  • Handle missing data and inconsistencies.
• Model Design & Training
• Select a feed-forward neural network or another architecture (CNN/RNN) and justify the choice.
• Experiment with different hyperparameters (e.g., number of layers, activation functions, learning rates).
• Use 10-fold cross-validation to optimize model performance.
• Evaluation & Reality Check
  • Evaluate model performance using accuracy on both the training and test datasets.
  • The final trained model will be tested on a hidden dataset to assess real-world generalization.
  • The best model and processed datasets must be submitted.

Files

• main.pdf: This contains the documentation that was created using knitr from the R Markdown file.
• main.Rmd: The final file where the model was created.
• all_models_details.txt: In this file, I logged models upon creation. Here, only promising models were logged, not every model tried.
• model_outputs: This folder contains the learning curve plots of the models in the model log file. The numbers correspond to the numbers in the log file.
• final_model_plot.png: The learning curve of the final model during training.
• reality_check.R: The file for the Reality Check. Upon execution, you will be prompted to select the dataset to run it.
• normalization_params.csv: This file contains the normalization parameters obtained from our training dataset and used for the normalization.
• model.keras: Our final Neural Network model with the best performance.