Artificial neural networks as adjuncts for assessing medical students' problem solving performances on computer-based...

Area of Science:

• Artificial Intelligence
• Immunology Education
• Educational Technology

Background:

• Computer-based simulations are valuable tools for immunology education.
• Understanding student problem-solving strategies is crucial for effective learning.
• Traditional assessment methods may not fully capture the nuances of complex problem-solving.

Purpose of the Study:

• To train artificial neural networks (ANNs) to recognize successful problem-solving patterns in immunology simulations.
• To classify student performance as successful or unsuccessful based on test selection.
• To analyze ANN outputs to understand student strategies and identify learning difficulties.

Main Methods:

• Supervised learning was used to train ANNs on student test selection data from seven immunology simulations.
• Trained ANNs evaluated new test selection patterns, classifying them as successful or unsuccessful.
• ANN output weights were analyzed to interpret problem-solving processes and identify distinct error patterns.

Main Results:

• ANNs correctly classified successful/unsuccessful solutions with over 90% accuracy.
• Successful solutions showed a progressive increase in ANN weights for relevant tests, indicating knowledge accumulation.
• Unsuccessful solutions were categorized into two patterns: extensive searching with poor recognition, or misrepresentation of problems.

Conclusions:

• ANNs can effectively model and predict student success in complex simulations.
• Analysis of ANN weights provides insights into cognitive processes during problem-solving.
• This AI-driven approach can identify specific learning difficulties and hypothesis-switching behaviors in students.