Figure 3
The diagram shows two neural network architectures displayed side by side. The left flowchart is labeled “G R U- R N N model predicting cheating from student exam results”. The sequence model begins with a box labeled “2-Layer G R U”. A downward arrow leads to an input labeled “G R U units (64, 32)”, with the text outside the box reading “Input Sequence” and “Sequence length T”. A downward arrow leads to “Dropout equals 0.2” and “Recurrent dropout equals 0.2”. A downward arrow leads to “Dense (16) plus Re L U”, with the text outside the box reading “d-dimensional features” along with an icon of a pile of documents. This flows into a block labeled “Output Dense (1) plus Sigmoid”, with an icon of sigma and integration symbols. A downward arrow leads to the last box labeled “R N N s model order-sensitive trajectories in student performance and capture temporal motifs”, accompanied by a small bar chart icon. The right flowchart is labeled “1 D-C N N model predicting cheating from student exam results”. The model begins with a box labeled “1 D-C N N”. A downward arrow leads to a block labeled “Conv 1 D (filters equals 32, kernel underscore size equals 3, padding equals ‘sam’)”, with a right arrow pointing to a text on the right reading “Burst patterns”. An icon of a connected network diagram is shown on the left. A downward arrow leads to the third block labeled “Conv 1 D (filters equals 64, kernel underscore size equals 5, padding equals ‘sam’)”, with a right arrow pointing to a text on the right reading “Shift invariance”. An icon of a connected network diagram is shown on the left. A downward arrow leads to the next block labeled “Dropout equals 0.3”. A downward arrow then leads to a block labeled “Output Dense (1) plus Sigmoid”, accompanied by an icon of sigma and integration symbols. A final downward arrow leads to a box labeled “1 D-C N N s detect localized temporal patterns”, accompanied by a small bar chart icon.GNU-RNN and 1D-CNN diagrams