Using Convolutional Neural Networks for Smart Classroom Observation
Classroom observation is one the primary tools for providing teachers feedback on their classroom practices. However, classroom observation is expensive and may suffer from bias. A partial automation of this process should go a long way towards improving the quantity and quality of feedback provided to a teacher that may eventually lead to better education. This paper presents the design and implementation of a Convolution Neural Network (CNN) that used class observation audio data to classify classroom activities into those based on the Stallings Classroom Snapshot. Data consisting of 859 audio episodes taken from a variety of semi-rural schools was used to train a CNN to classify the audio episode into one of six Stalling categories; Classwork, Classroom Management, Lecture/Demonstration, Practice Drill, Discussion/Question and Answer, and Reading Aloud. The resulting model yielded a best accuracy of 89.97%. The model performed better than previous model built using KNN and Random Forest. The 20 MB model implementation of the CNN on an iPhone SE using CoreML yielded an average response time of 0.8 seconds/inference (sd = 0.01).
Khan, Muhammed S. and Zualkernan, Imran, "Using Convolutional Neural Networks for Smart Classroom Observation" (2020). Regis University Faculty Publications. 147.