Predictive Modeling and Visualization

Overview

This repository contains R scripts and analyses for predictive modeling and data visualization, developed as part of coursework for STAT 561 at SIUE. The projects cover a range of statistical and machine learning techniques applied to diverse datasets, including absenteeism at work, customer churn, patient satisfaction, muscle mass, county demographic information (CDI), flu shot uptake, student performance, and loan approval prediction. The analyses include regression models, regularization techniques, and diagnostic visualizations to explore relationships and predict outcomes. Topics Covered The repository encompasses the following topics in predictive modeling and visualization:

Link to the dataset:

Iranian Churn: https://archive.ics.uci.edu/dataset/563/iranian+churn+dataset

Patient satisfaction: https://github.com/Prashanna-Raj-Pandit/STAT-Predictive-Modeling/blob/main/Linear%20Regression%20Model/pat_stat.txt

Absenteeism at work: https://github.com/Prashanna-Raj-Pandit/STAT-Predictive-Modeling/blob/main/Logistic%20Regression/Absenteeism_at_work.xls

Flu shot: https://github.com/Prashanna-Raj-Pandit/STAT-Predictive-Modeling/blob/main/Logistic%20Regression/flu%20shot.txt

Student Performance: https://github.com/Prashanna-Raj-Pandit/STAT-Predictive-Modeling/blob/main/Model%20diagnostic/Student_Performance.csv

Loan Data: https://github.com/Prashanna-Raj-Pandit/STAT-Predictive-Modeling/blob/main/Model%20diagnostic/loan_data.csv

muscle mass: https://github.com/Prashanna-Raj-Pandit/STAT-Predictive-Modeling/blob/main/Linear%20Regression%20Model/muscle_mass.txt

Iranian Churn Analysis (Tableau):

Data Preprocessing:

• Handling missing values (e.g., imputation in loan data)
• Converting variables to factors for categorical analysis
• Data splitting into training, test, and holdout sets
• Recoding variables (e.g., absenteeism categories)

Exploratory Data Analysis (EDA):

• Histograms, boxplots, and scatter plots for distribution and relationship analysis
• Correlation matrices to assess relationships between numeric variables Contingency tables and bar plots for categorical variables

Regression Models:

• Multiple Linear Regression: Predicting continuous outcomes (e.g., student performance, patient satisfaction)
• Poisson Regression: Modeling count data (e.g., frequency of use in customer churn)
• Negative Binomial Regression: Addressing overdispersion in count data
• Logistic Regression: Binary classification (e.g., loan approval, flu shot uptake)
• Multinomial Logistic Regression: Multi-class classification (e.g., absenteeism categories)

Regularization Techniques:

• Lasso Regression: Feature selection and shrinkage (e.g., absenteeism, customer churn)
• Ridge Regression: Handling multicollinearity (e.g., absenteeism, customer churn)
• Elastic Net Regression: Combining Lasso and Ridge with tuned alpha and lambda parameters

Model Selection:

• Forward and backward stepwise selection for variable selection
• Cross-validation (e.g., 5-fold, 10-fold) to optimize model parameters

Model Diagnostics:

• Residual plots (vs. fitted values, predictors, observation order) to check linearity, homoscedasticity, and independence
• Q-Q plots and histograms for normality of residuals
• Variance Inflation Factor (VIF) for multicollinearity checks
• Overdispersion checks for Poisson models

Model Evaluation:

• Metrics: RMSE, R-squared, accuracy, sensitivity, specificity, AUC, and confusion matrices
• ROC curves for binary classification models
• Comparison of model performance across different techniques

Visualization:

• Scatter plot matrices (pairs plots)
• Residual diagnostic plots
• Bar plots and stacked bar plots for categorical data
• Regression line and curve fitting (e.g., linear, quadratic, cubic models for muscle mass)

Applications:

• Predicting absenteeism time and categories in workplace data
• Modeling customer churn and frequency of use in telecom data
• Analyzing patient satisfaction based on age, severity, and anxiety
• Studying muscle mass decline with age using polynomial regression
• Predicting physician counts based on demographic and regional factors
• Modeling flu shot uptake based on age, awareness, and sex
• Predicting student performance using academic preparation factors
• Predicting loan approval based on applicant characteristics

Repository Structure

LinearRegression.R: Linear regression models for patient satisfaction, muscle mass (with polynomial terms), and physician counts, including forward/backward selection.

logistics_regression.R: Multinomial logistic regression for absenteeism categories and binary logistic regression for flu shot uptake, with EDA and model selection.

Model diagnostics.pdf: Analysis of student performance (linear regression) and loan approval (logistic regression), with detailed model diagnostics and proposed interventions.

Regularization.R: Analysis of absenteeism data using Lasso, Ridge, and Elastic Net regression for continuous outcomes and Lasso for multinomial classification.

Project.R: Customer churn analysis with Poisson, Negative Binomial, Ridge, and Lasso regression, plus logistic regression for complaint prediction with SMOTE for class imbalance.

Usage

Each script is self-contained and can be run independently in R or RStudio. Ensure the required datasets (e.g., Absenteeism_at_work.xls, Customer Churn.csv, pat_stat.txt, muscle_mass.txt, cdi.txt, flu shot.txt, Student_Performance.csv, loan_data.csv) are in the working directory or update the file paths in the scripts. To run a script:

Open the script in R or RStudio (e.g., HW4.R).

Ensure the required dataset is available. Run the script to perform the analysis and generate outputs (e.g., model summaries, plots, performance metrics).

Example: source("HW4.R")

This will load the absenteeism dataset, preprocess it, fit regression models, and output test MSEs and classification errors. Datasets

Example Outputs

• Absenteeism Analysis (HW4.R): Compares Lasso, Ridge, and Elastic Net models for predicting absenteeism hours (Test MSEs) and Lasso for classifying absenteeism categories (classification error).
• Customer Churn (Project.R): Compares Poisson, Negative Binomial, Ridge, and Lasso models for frequency of use (RMSE) and logistic regression for complaints (AUC, confusion matrix).
• Patient Satisfaction (LinearRegression.R): Linear regression model with forward/backward selection, predicting satisfaction based on age, severity, and anxiety.
• Muscle Mass (LinearRegression.R): Linear, quadratic, and cubic models for muscle mass vs. age, with visualization of regression fits.
• Flu Shot (logistics_regression.R): Logistic regression for flu shot uptake, with forward/backward selection and predicted probabilities.
• Student Performance (Model diagnostics.pdf): Linear regression with cross-validation, achieving an R-squared of 0.9893 and RMSE of 1.99, with residual diagnostics confirming model assumptions.
• Loan Approval (Model diagnostics.pdf): Logistic regression with 85.53% accuracy, 100% sensitivity, and 50% specificity, with ROC curve analysis.

Reference: Iranian Churn [Dataset]. (2020). UCI Machine Learning Repository. https://doi.org/10.24432/C5JW3Z.