Dry Beans Multiclass Classification
ML models built from scratch to classify 7 types of dry beans based on physical characteristics
Project Overview
Built classification models from scratch to accurately identify and classify 7 types of dry beans based on their physical characteristics and properties. This project demonstrates deep understanding of machine learning fundamentals by implementing algorithms without using high-level libraries.
The Challenge
Dry beans are an important food source worldwide, and accurate classification is crucial for agricultural quality control and pricing. The goal was to develop models that could automatically classify beans based on features like shape, size, and other physical properties – all while implementing the algorithms from scratch to truly understand their mechanics.
Implementation
Developed three classification algorithms completely from scratch without Scikit- Learn:
-
Logistic Regression – Implemented multiclass classification using one-vs-all approach with gradient descent optimization
-
Support Vector Machines (SVM) – Built SVM classifier with kernel methods for non-linear decision boundaries
-
Neural Networks – Created a fully-connected neural network with backpropagation for deep learning classification
Dataset
Used the UC Irvine Machine Learning Repository’s Dry Bean Dataset containing:
- 13,611 samples of dry beans
- 16 features (12 dimensional + 4 shape features)
- 7 distinct classes of beans
My Contribution
We worked on this project as a group of 3 as part of the Statistical Learning in Engineering course [ IE 7300 ]. I took the lead on comprehensive technical documentation and analysis. My work involved:
- Created a comprehensive technical report explaining how all three models (Logistic Regression, SVM, and Neural Networks) work and their implementation details
- Documented key findings from exploratory data analysis, including feature correlations and class imbalance issues
- Provided strategic recommendations on model selection based on the accuracy-speed tradeoff analysis
The report serves as a complete reference for understanding the project’s approach and results.