Machine Learning Exercises -- EMCH 368

A collection of four MATLAB machine learning exercises covering regression, text classification, digit recognition, and image classification. Each exercise includes datasets, training scripts, and pre-trained models so results can be reproduced without retraining.

Requirements

MATLAB R2023a or later with the following toolboxes installed:

Toolbox	Used In
Statistics and Machine Learning Toolbox	Exercises 1, 2
Deep Learning Toolbox	Exercises 3, 4
Image Processing Toolbox	Exercises 3, 4

Run check_libraries.m from the repository root to verify your installation.

Exercises

Exercise 1 -- Linear Regression (Housing Prices)

Implements gradient descent from scratch to fit a polynomial regression surface predicting home sale prices from square footage and year built.

cd Exercise_1_regression
Machine_Learning_1a_home_data

Dataset: train.csv (Ames Housing data subset)

Exercise 2 -- Text Sentiment Analysis (Random Forest)

Trains a 1000-tree Random Forest on pre-computed text embeddings to classify review sentiment. Includes OOB error visualization during training.

cd Exercise_2_Text_Sentience
Text_sentience

Set trainModel = true on line 7 to retrain from scratch, or leave it false to load the included pre-trained model.

Dataset: train_data.parquet, test_data.parquet

Exercise 3 -- Digit Classification (Neural Network)

Trains a three-layer pattern recognition network (300-200-100 hidden units) on the MNIST dataset.

cd Exercise_3_Digit_Classification
MNIST_ML_Lab9

Set trainModel = true on line 12 to retrain, or leave it false to load the included model.

Dataset: mnist.mat

Exercise 4 -- Image Classification (Residual Network)

Trains a ResNet on CIFAR-10 with data augmentation (random reflections, translations) for 80 epochs. Displays a confusion matrix and sample predictions on the validation set.

cd Exercise_4_Image_Classification
CIFAR_10

If trainedResidualNetwork.mat is not present, the script will train automatically. Otherwise it loads the saved network.

Dataset: CIFAR-10 batches in data/

Repository Structure

.
├── check_libraries.m              # Verify required toolboxes
├── Library_Check/
│   └── checkLibrary.m             # Alternate toolbox checker
├── Exercise_1_regression/
│   ├── Machine_Learning_1a_home_data.m
│   ├── preprocessAndSplitData.m
│   └── train.csv
├── Exercise_2_Text_Sentience/
│   ├── Text_sentience.m
│   ├── train_data.parquet
│   ├── test_data.parquet
│   └── trainedModel.mat
├── Exercise_3_Digit_Classification/
│   ├── MNIST_ML_Lab9.m
│   ├── mnist.mat
│   └── trainedModel.mat
└── Exercise_4_Image_Classification/
    ├── CIFAR_10.m
    ├── loadCIFARData.m
    ├── trainedResidualNetwork.mat
    └── data/
        ├── batches.meta.mat
        ├── data_batch_[1-5].mat
        └── test_batch.mat

License

This project is licensed under the MIT License. See LICENSE for details.