Assessing Children's Fine Motor Skills With Sensor-Augmented Toys: Machine Learning Approach.

Area of Science:

• Pediatric assessment
• Motor development
• Human-computer interaction

Background:

• 5%-10% of elementary students exhibit delayed fine motor skills.
• Current assessment tools are time-consuming and lack child engagement.
• Sensor-augmented toys and machine learning offer potential solutions.

Purpose of the Study:

• To evaluate sensor-augmented toys for assessing children's fine motor skills.
• To predict outcomes of the Movement Assessment Battery for Children Second Edition (fine MABC-2).
• To analyze the impact of game type, data type, and difficulty on prediction accuracy.

Main Methods:

• 95 elementary school children (mean age 7.8 years) used the Futuro Cube toy.
• Children played two games ('roadrunner' for speed, 'maze' for precision) with varying difficulty levels.
• Four machine learning classifiers (KNN, LR, DT, SVM) analyzed sensor and game data to predict fine MABC-2 scores.

Main Results:

• The Decision Tree (DT) classifier achieved the highest accuracy (0.76) using sensor and game data from the 'roadrunner' game (easiest and hardest levels).
• Game type significantly influenced accuracy across all classifiers (P<.05).
• Classifier type and difficulty level had less impact on prediction accuracy.

Conclusions:

• Sensor-augmented toys provide an efficient method for predicting fine motor skill scores in elementary school children.
• Game type (speed vs. precision) and data type are more critical than classifier choice or difficulty level for accurate assessment.
• This technology offers a more engaging and potentially scalable approach to fine motor skill evaluation.