Figure 10
The figure contains six line plots arranged in three rows and two columns, labeled (a) through (f). All plots use “Epoch” on the horizontal axis. The left-column plots show mean squared error with the vertical axis labeled “M S E”, and the right-column plots show coefficient of determination with the vertical axis labeled “R-squared”. Each plot in the left column includes two curves: a line with the square markers labeled “Train loss” and a line with the circular markers labeled “Validation loss”. Each plot in the right column includes two curves: a line with the square markers labeled “Train R-squared” and a line with the circular markers labeled “Validation R-squared”. Light dashed gridlines appear in the background of all panels. Panel (a), titled “Loss curve of lightweight model”: The horizontal axis ranges from 0 to 120 with an interval of 20, and the vertical axis ranges from 0.00 to 0.16 with an interval of 0.02. It shows training M S E, starting from 0.148, and validation M S E, starting from 0.068, decreasing rapidly during the early epochs and then gradually leveling off at low values by around 120 epochs. The validation loss follows a similar trend with small fluctuations. Panel (b), titled “R-squared curve of lightweight model”: The horizontal axis ranges from 0 to 120 with an interval of 20, and the vertical axis ranges from negative 10 to 2 with an interval of 2. It shows training R-squared, starting from negative 8.3, and validation R-squared, starting from negative 1, increasing at early epochs toward values near 1 as epochs increase, with validation R-squared showing slight oscillations. Panel (c), titled “Loss curve of Model-1”: The horizontal axis ranges from 0 to 100 with an interval of 20, and the vertical axis ranges from 0.00 to 0.07 with an interval of 0.01. It shows training M S E, starting from 0.07, and validation M S E, starting from 0.015, steadily decreasing toward near-zero values by about 100 epochs. Panel (d), titled “R-squared curve of Model-1”: The horizontal axis ranges from 0 to 100 with an interval of 20, and the vertical axis ranges from negative 6 to 2 with an interval of 2. It shows R-squared rising from negative values toward positive values close to 1, with validation R-squared fluctuating slightly above the training curve. Panel (e), titled “Loss curve of Model-2”: The horizontal axis ranges from 0 to 120 with an interval of 20, and the vertical axis ranges from 0.00 to 0.06 with an interval of 0.01. It shows training M S E, starting from 0.055, and validation M S E, starting from 0.005, decreasing quickly at first and then tapering to very small values by around 120 epochs, with validation loss remaining consistently low. Panel (f), titled “R-squared curve of Model-2”: The horizontal axis ranges from 0 to 120 with an interval of 20, and the vertical axis ranges from negative 5 to 2 with an interval of 1. It shows training and validation R-squared increasing from negative 5 and 0, respectively, and converging toward values close to 1 as training progresses. Note: All the numerical data values are approximated.Architecture of the enhanced CNN model used for predicting local property ratios