This repository is a curated collection of my work and solutions to case studies, mini-projects, and a guided capstone that are part of the Springboard Data Science program.
This README contains summaries (objectives, methods, and takeaways) of the following curriculum categories:
- Case Studies: Industry case studies that cover a wide range of topics and techniques in data science
- Statistical Foundations: Case studies that explore and apply fundamental statistical techniques, without necessarily building predictive models
- Mini Projects: Projects using industry-standard tools and libraries for data manipulation and analysis
- Guided Capstone: An end-to-end data science project experience
Not included in this repo are my two independent capstones, which have their own repositories. Links below:
For more of my project work or independent capstones, please visit the pinned repositories on my Github or visit my website: https://www.dennisvdang.com.
-
Case Studies
- Customer Segmentation (Clustering Case Study)
- Customer Purchasing Prediction (Decision Tree Case Study)
- COVID-19 Prediction (Random Forest Case Study)
- Diabetes Prediction (Grid Search KNN Case Study)
- London Housing Price Forecasting (Time Series Analysis/EDA Case Study)
- Wine Quality Prediction (Linear Regression Case Study)
- Heart Disease Prediction (Logistic Regression Case Study)
- Passenger Survival Prediction (Gradient Boosting Case Study)
- Flight Delay Prediction (Bayesian Optimization Case Study)
- Sales Forecasting (Time Series Forecasting with ARIMA Case Study)
-
Statistical Foundations
-
Mini Projects
-
Guided Capstone
Below are summaries of case studies that cover a wide range of topics and demonstrate the application of various data science techniques and algorithms.
Objective
- Segment customers based on their purchasing behavior and response to marketing campaign offers, and provide insights into the characteristics of each customer segment to enable targeted marketing strategies
Methodology
- Cleaned and transformed data to create a matrix of customer responses to different offers
- Applied k-means clustering and determined the optimal number of clusters using the Elbow, Silhouette, and Gap statistic methods
- Visualized the customer segments in a 2D space using Principal Component Analysis (PCA)
- Analyzed the distribution of key features across the identified clusters and revealed the characteristics of each customer segment
- Libraries:
Pandas,Scikit-learn,Matplotlib,Seaborn
Key Takeaways
- Identified distinct customer segments with varying offer preferences and purchasing behaviors
- Demonstrated the effectiveness of k-means clustering for customer segmentation
- Highlighted the importance of feature selection and dimensionality reduction for effective clustering and visualization
Link to Customer Segmentation Case Study notebook
Objective
- Build a predictive model to estimate the likelihood of RR Diner Coffee's loyal customers purchasing the Hidden Farm coffee
Methodology
- Analyzed customer data such as age, gender, salary, online purchases, distance from the flagship store, and spending habits to understand key drivers of customer purchasing behavior
- Implemented four decision tree models with varying parameters to understand the impact of different parameters on model performance:
- Default Decision Tree Model
- Decision Tree with Max Depth = 3
- Decision Tree with Min Samples Split = 10
- Decision Tree with Min Samples Leaf = 5
- Evaluated model performance using accuracy, balanced accuracy, precision, and recall metrics
- Compared the performance of a single decision tree model to a Random Forest ensemble model
- Libraries:
Pandas,Scikit-learn,Matplotlib
Key Takeaways
- Identified key factors influencing customer purchasing decisions
- Demonstrated the effectiveness of decision trees for predictive modeling
- Highlighted the importance of model simplicity and interpretability for specific datasets
Link to Customer Purchasing Prediction Case Study notebook
Objective
- Predict the severity of COVID-19 cases using patient characteristics and symptoms
Methodology
- Preprocessed and explored the dataset to understand feature-target relationships
- Built and evaluated random forest models with different hyperparameters and feature subsets
- Interpreted feature importances to identify key risk factors for severe COVID-19
- Libraries:
Pandas,Numpy,Scikit-learn,Matplotlib,Seaborn
Key Takeaways
- Demonstrated the application of random forests for predicting disease severity
- Highlighted the importance of feature selection and hyperparameter tuning
Link to COVID-19 Random Forest Case Study notebook
Objective
- Optimize the hyperparameters of a K-Nearest Neighbors (KNN) classifier using grid search on the Pima Indians Diabetes dataset.
Methodology
- Loaded and preprocessed the Pima Indians Diabetes dataset and standardize the features.
- Implemented a loop to train KNN models with a range of neighbor values (1-9) and evaluated the training and test scores.
- Identified the optimal number of neighbors based on the maximum test score.
- Fit the KNN model with the optimal number of neighbors and evaluated its performance on the training and test sets.
- Plotted the confusion matrix and printed the classification report for the optimal model.
- Applied the grid search method to find the optimal number of estimators in a Random Forest model.
- Libraries:
Pandas,Scikit-learn,Matplotlib
Key Takeaways
- Grid search is an effective technique for optimizing hyperparameters of machine learning models, such as the number of neighbors in KNN and the number of estimators in Random Forest.
- Identifying the optimal hyperparameters can significantly improve the model's performance on both the training and test datasets.
- Visualizing the model's performance, such as through confusion matrices and classification reports, provides valuable insights into the model's strengths and weaknesses.
Link to Diabetes Prediction Case Study notebook
Objective
- Analyze trends in London's housing market and forecast future prices
Methodology
- Conducted exploratory data analysis (EDA) to understand the impact of the 2008 financial crisis on London's housing market
- Applied statistical tests to explore the relationship between pre-crisis prices and crisis recovery time
- Employed time series models (ARIMA, SARIMA) and Random Forest to forecast property values
- Evaluated model performance using MSE and MAE
- Libraries:
Pandas,Numpy,Statsmodels,Matplotlib,Seaborn
Key Takeaways
- Identified trends and patterns in London's housing market over time
- Demonstrated the application of time series modeling for forecasting
Link to London Housing Case Study notebook
Objective
- Predict alcohol levels in wine using regression analysis and build an accurate model
Methodology
- Performed univariate and multivariate analysis to iterate towards an accurate model
- Utilized EDA to visualize correlations and inform model selection
- Developed multiple linear regression models, iterating to improve accuracy and reduce redundancy
- Evaluated models based on R-squared, AIC, and BIC metrics
- Libraries:
Pandas,NumPy,Matplotlib,Seaborn,Statsmodels,Scikit-learn
Key Takeaways
- Importance of iterative model development and feature selection
- Demonstrated the application of linear regression in predicting continuous variables
- Identified the most elegant and economical model (rModel4) using a few predictors effectively
Link to Linear Regression Case Study notebook
Objective
- Build a logistic regression model to predict the presence of heart disease using patient health data
Methodology
- Introduced logistic regression as a fundamental classification algorithm
- Covered key concepts like classification, model evaluation, discriminative vs generative classifiers
- Performed data preprocessing (handling missing values, encoding categorical variables)
- Split the data into training and testing sets
- Built and evaluated logistic regression models using scikit-learn
- Explored model tuning and evaluation metrics (accuracy, precision, recall, ROC curves)
- Visualized decision boundaries and interpreted model coefficients
- Libraries:
Pandas,Scikit-learn,Matplotlib,NumPy
Key Takeaways
- Logistic regression is a simple yet effective algorithm for binary classification problems
- Importance of proper data preprocessing and model evaluation for classification tasks
- Logistic regression provides interpretable models and discriminative decision boundaries
- Need to consider multiple evaluation metrics beyond accuracy for imbalanced datasets
Link to Logistic Regression Case Study notebook
Objective
- Explore the concept of gradient boosting using decision trees as base predictors to predict passenger survival in the Titanic dataset
Methodology
- Utilized the Titanic dataset to demonstrate the application of gradient boosting for binary classification
- Preprocessed the dataset, handling missing values and encoding categorical variables
- Demonstrated the process of fitting a series of decision trees on residual errors to improve predictions
- Evaluated model performance with multiple learning rates and calculated the ROC curve
- Libraries:
Pandas,NumPy,Scikit-learn,Matplotlib
Key Takeaways
- Gradient boosting can significantly improve model performance in predicting passenger survival
- Importance of tuning hyperparameters, such as learning rate and number of estimators, to optimize model performance
- Demonstrated the effectiveness of gradient boosting in handling complex datasets with a mix of numerical and categorical features
Link to Gradient Boosting Case Study notebook
Objective
- Use Bayesian Optimization to tune hyperparameters of a LightGBM model for flight delay prediction
Methodology
- Loaded and preprocessed the flight delay dataset, handling missing values and encoding categorical features.
- Evaluated the LightGBM model with given hyperparameters using cross-validation.
- Set up the Bayesian Optimization process using the
BayesianOptimizationlibrary, specifying the hyperparameter search space. - Performed Bayesian Optimization to find the optimal hyperparameters that maximize the AUC (Area Under the Receiver Operating Characteristic Curve) metric.
- Trained the final LightGBM model with the optimal hyperparameters obtained from Bayesian Optimization.
- Evaluated the final model's performance on the test set and compared it with a majority class baseline.
- Libraries:
Pandas,NumPy,LightGBM,Scikit-learn,Bayesian Optimization,Matplotlib
Key Takeaways
- Bayesian Optimization is an effective technique for optimizing hyperparameters of complex machine learning models, such as LightGBM.
- Visualizing the model's performance using metrics like AUC, confusion matrix, and accuracy provides insights into the model's strengths and weaknesses.
Link to Bayesian Optimization Case Study notebook
Objective
- To analyze the sales data of Cowboy Cigarettes from 1949 to 1960 and predict future sales using time series forecasting with the ARIMA model.
Methodology
- Loaded and explored the historical sales data of Cowboy Cigarettes.
- Performed data cleaning and transformation, including log transformation to stabilize the variance.
- Conducted exploratory data analysis (EDA) to understand the trend, seasonality, and noise in the sales data.
- Utilized the Augmented Dickey-Fuller test to check for stationarity and applied differencing to achieve a stationary time series.
- Employed the ARIMA model for time series forecasting, optimizing the model parameters (p, d, q) based on the Akaike Information Criterion (AIC).
- Forecasted future sales for the next two years and visualized the original data along with the forecasts.
Key Takeaways:
- The ARIMA model effectively captured the underlying trend and seasonality in Cowboy Cigarettes' sales data, providing a close fit to the historical data.
- Forecasting revealed an increasing trend in sales, suggesting a growing demand for Cowboy Cigarettes.
- The case study highlighted the importance of time series decomposition and the capability of ARIMA models in forecasting future trends based on historical data.
- Libraries:
Pandas,NumPy,Matplotlib,Seaborn,Statsmodels
Link to Time Series Analysis and Forecasting with ARIMA Case Study notebook
Summaries of case studies that explore and apply fundamental statistical techniques and mathematical principles, without necessarily building predictive models.
Objective
- Use cosine similarity to compare numeric data and text datasets
Methodology
- Calculated similarity measures for sentences/paragraphs
- Libraries:
NumPy,Pandas,Matplotlib,Scipy,Scikit-learn
Key Takeaways
- Cosine similarity effective for comparing numeric and text data
- Applicable in NLP and recommendation systems
Link to Cosine Similarity Case Study notebook
Objective
- Demonstrate calculation and comparison of Euclidean and Manhattan distances
Methodology
- Visualized distribution of distances through histograms
- Highlighted application in data science, foundational for PCA
- Libraries:
Pandas,NumPy,Matplotlib
Key Takeaways
- Euclidean and Manhattan distances have different behaviors
Link to Euclidean and Manhattan Distance Case Study notebook
Objective
- Apply Frequentist inference to real-world data, addressing practical business questions for a hospital
Methodology
- Emphasized statistical concepts like z-statistic, t-statistic, Central Limit Theorem, and confidence intervals
- Explored the Central Limit Theorem and its implications
- Estimated population mean and standard deviation from a sample
- Calculated confidence intervals
- Libraries:
Pandas,NumPy,Matplotlib
Key Takeaways
- Demonstrated the application of Frequentist inference in making data-driven decisions
- Highlighted the importance of statistical concepts in real-world scenarios
Link to Frequentist Inference Case Study notebook (Part 1) (Part 2)
Objective
- Explore the use of APIs for data collection and integration in data analysis workflows
Methodology
- Demonstrated making API requests, handling authentication, and parsing JSON responses
- Collected data from the GitHub API and performed basic analysis
- Libraries:
Requests,Pandas,Matplotlib
Key Takeaways
- APIs enable access to vast amounts of data for analysis
- Python libraries simplify API interactions and data manipulation
Objective
- Demonstrate the use of SQL for data manipulation, aggregation, and analysis
Methodology
- Performed various SQL queries on a sample database (filtering, joining, grouping, sorting)
- Showcased SQL's power in handling complex data queries and transformations
- Libraries/Tools:
SQLite,Pandas,MySQL,PHPMyAdmin
Key Takeaways
- SQL is a powerful tool for data manipulation and analysis
- Proficiency in SQL and SQL libraries in Python is essential for working with relational databases
Objective
- Complete a comprehensive data science project covering all stages of the data science lifecycle
Methodology
- Collaborated with a team to define project scope, objectives, and deliverables
- Conducted extensive EDA to gain insights and inform modeling
- Built and evaluated multiple machine learning models (regression, classification, clustering)
- Communicated findings through visualizations, reports, and presentations
- Libraries:
Pandas,Numpy,Scikit-learn,Matplotlib,Seaborn
Key Takeaways
- Hands-on experience with the complete data science lifecycle
- Importance of collaboration, communication, and project management in data science