AutoMixReg

AutoMixReg is an R package for simulating and fitting Mixture of Linear Regressions using the Expectation-Maximization (EM) algorithm. It includes functionality for:

• Generating synthetic data from known mixture models
• Fitting mixture regression models using EM
• Automatically selecting the optimal number of components using Bayesian Information Criterion (BIC)
• Running a full model selection and fitting pipeline in a single line

This package is useful for unsupervised regression modeling, model-based clustering, and statistical learning in high-dimensional settings.

📦 Installation

# Install devtools if not already installed
install.packages("devtools")

# Install AutoMixReg from GitHub
devtools::install_github("Fadhaa/AutoMixReg")

Example Usage

1. Simulate Data from a Mixture of Linear Regressions

library(AutoMixReg)

custom_betas <- list(
  c(0.5, 1.5, 2, 3),   # Cluster 1
  c(1, 2, 3, 4),       # Cluster 2
  c(2, 3, 4, 6)        # Cluster 3
)

df <- generate_mixture_data(
  n_samples = 1000,
  n_features = 3,
  betas = custom_betas,
  cluster_probs = c(0.4, 0.3, 0.3),
  noise_std = 1,
  seed = 42
)

head(df)

2. Fit a Mixture Regression Model (EM Algorithm)

feature_cols <- setdiff(names(df), c("y", "cluster"))
X <- cbind(1, as.matrix(df[, feature_cols]))
y <- df$y

model <- fit_mixture_regression(X, y, n_components = 3)

# Inspect estimated parameters
model$weights
model$betas
model$sigmas

3. Automatically Select Best k Using BIC

bic_result <- select_best_k_bic(X, y, k_max = 6)
bic_result$best_k

4. One-Line Run Pipeline

run_mixture_pipeline(df)