Mid-Term Project: Stroke Risk Classification
Project Overview
In this mid-term project, you will work with a real-world healthcare dataset focused on stroke risk classification.
The dataset contains features related to patients’ medical history and health conditions. Your task is to preprocess
the data, address common issues such as missing values, irrelevant and redundant features, and perform
classification using machine learning models.
This project will assess your understanding of data preprocessing, dimensionality reduction, handling class
imbalance, and applying classification algorithms. You'll also employ K-Fold Cross-Validation to ensure the
robustness of your models.
Part 1: Data Preprocessing and Exploration
1. Irrelevant Features
Task: Analyze the dataset and identify any irrelevant features.
Questions:
- Which features have little to no contribution to stroke risk prediction, and what are the reasons for their lack of
significance?
2. Missing Values
Task: Investigate the dataset to identify missing values.
Questions:
- Does the dataset contain missing values? Which attributes have missing values?
- What strategy did you use to handle missing data (e.g., imputation, removal), and why?
3. Noisy Data
Task: Analyze the dataset for noisy data points or outliers.
Questions:
- Does the dataset contain noisy data or outliers? How did you detect noise in the data?
- Which features are affected by noise, and how did you address it?
- Justify the approach you used to clean the noisy data.
4. Redundant Features
Task: Check if the dataset contains redundant features.
Questions:
- Are there any features that provide similar information (e.g., highly correlated features)?
- What method did you use to detect redundancy?
- How did you handle the redundant features? Provide reasoning for your approach.
5. Dimensionality Reduction
Task: Determine whether dimensionality reduction is necessary for this dataset.
Questions:
- Do you need to reduce the dimensionality of the dataset? Why or why not?
- If applicable, which technique did you choose for dimensionality reduction (e.g., PCA, Feature Selection)? Explain
your choice.
- How many dimensions did you reduce the dataset to, and why did you choose that number?
6. Class Imbalance
Task: Analyze the class distribution to check for imbalanced data.
Questions:
- Is there a class imbalance in the dataset (e.g., more 'No Risk' patients than 'High Risk')?
- What technique did you use to handle class imbalance (e.g., oversampling, SMOTE)? Justify your choice.
Part 2: Model Building and Classification
7. K-Fold Cross-Validation
Task: Implement K-Fold Cross-Validation to evaluate the models.
Questions:
- How did you choose the value of k, and what effect does it have on the validation process?
8. Model Selection and Training
Task: Train two different classification models using the training dataset.
Options: Choose two algorithms from the following list:
- Decision Tree
- k-Nearest Neighbors (k-NN)
- Naive Bayes
- Random Forest
- XGBoost
Questions:
- Which two algorithms did you choose, and why?
- Provide a theoretical justification for your choices, considering factors such as time complexity, accuracy,
interpretability, and the nature of the dataset.
- Explain how you selected the model parameters.
9. Model Evaluation
Task: Evaluate the performance of each model using appropriate metrics.
Metrics: Use the following evaluation metrics to assess each model:
- Accuracy
- Confusion Matrix
- Precision
- Recall
- F1-Score
Questions:
- Present the evaluation results for each model and compare their performance.
- Which model performed best in terms of accuracy? Discuss whether accuracy is the best metric to focus on for
this problem. If not, which alternative metric would be more appropriate, and why?
10. Model Comparison and Best Model
Task: Compare the results of the two models and determine the best-performing one.
Questions:
- Which model gave the best overall performance, and why? Discuss the trade-offs between time, accuracy,
interpretability, and generalizability.
- What are the key factors contributing to the success of the best-performing model?
Part 3: Final Report and Submission
Final Report:
Create a concise and well-structured project report that covers the following:
Data Preprocessing: Summarize the steps taken to clean and prepare the dataset, including how you handled
missing values, irrelevant and redundant features, noisy data, and any dimensionality reduction. Ensure that
each preprocessing decision is clearly explained, answering the questions posed in Part 1.
Model Selection and Training: Provide an overview of the models selected for classification, explaining why you
chose them and how they were trained, including any hyperparameter tuning or cross-validation techniques
used. Ensure that you address all questions from Part 2 related to model choices and training processes.
Results and Analysis: Present the performance of each model, including key evaluation metrics (e.g., accuracy,
precision, recall, F1-score) and a comparison between the models. Support your analysis with clear and concise
visualizations (e.g., confusion matrices, graphs).
Code Submission:
- Submit your code in a Jupyter Notebook, making sure it is well-organized and properly commented.
- The notebook should include all steps, from data preprocessing to model training, evaluation, and comparison.
- Ensure that the output results are clearly displayed in the notebook for each step.
Deliverables:
- Project report (PDF or Word format) containing detailed answers to all tasks.
- Code in Jupyter Notebook.