Novel transfer learning approach for hand drawn mathematical geometric shapes classification

Area of Science:

• Computer Science
• Mathematics Education

Background:

• Hand-drawn geometric shapes are crucial for understanding mathematical concepts.
• Accurate recognition of these shapes aids in digitizing notes and enhancing educational tools.

Purpose of the Study:

• To develop an innovative machine learning algorithm for automatic classification of hand-drawn mathematical geometric shapes.
• To facilitate the seamless integration of handwritten mathematical input with digital systems.

Main Methods:

• A novel algorithm, CnN-RFc, was developed, combining Convolutional Neural Networks (CNN) for spatial feature extraction and Random Forest (RF) for probabilistic feature extraction.
• A benchmark dataset of 20,000 images across eight geometric shape classes was utilized.

Main Results:

• The CnN-RFc method, utilizing the Light Gradient Boosting Machine (LGBM) algorithm, achieved a high accuracy score of 98% in classifying hand-drawn shapes.
• This performance surpasses existing state-of-the-art approaches for this task.

Conclusions:

• The proposed CnN-RFc algorithm offers a highly accurate solution for classifying hand-drawn mathematical geometric shapes.
• This technology has significant applications in educational platforms, enabling interactive learning and immediate student feedback.