This indicates that the weight values ALK mutation (within their respective ranges) have been distributed spatially among the prototype vectors, with the neighboring vectors having similar weights. Figure 3 Maps of weight components after SOM training. Table 2 Statistics of the weight values of the trained SOM. From the maps in Figure 3, it can be seen that the neurons at the lower left corner has low follower’s velocities, almost zero relative velocities (wxy2 value in the mid-range)

and small gaps. They represent the state where vehicles are queuing in congested conditions. In this condition, the follower is expected to accelerate or decelerate with small magnitudes. The neurons located at the top right corner of the grid represent stimulus with relatively high follower’s velocities, negative relative velocities (wxy2 less than midvalue), and large gaps. This condition indicates that the follower is closing in to the leader from a distance (but may not necessarily decelerate). The neurons at the top left corner have moderate follower’s velocities, high relative velocities, and moderate gaps. They represent the scenario that the lead vehicle is accelerating away from the follower. The follower may then respond by accelerating. The neurons at the bottom right corner have weight vectors that have moderately high follower’s velocities, negative relative velocities, and small gaps. These

prototype vectors represent the condition that the follower is quickly closing in to the leader. The driver of the following vehicle is likely to apply his/her brake. 5.2. Distribution of Mean Response For each neuron, the mean response (average follower’s acceleration) computed from the winning vectors is next plotted in Figure 4. Figure 4(a) shows the distribution of mean response calculated

from the training data set. For each x value in the map, as y increases from 0 to 10, the mean response changes from deceleration to acceleration. For each y value in the map, as x increases from 0 to 10, the mean response changes from acceleration to deceleration. The maximum acceleration occurs near x = 0, y = 10, which is the top left corner of the SOM as shown in Figure 3. On the other hand, the maximum deceleration occurs near x = 10, y = 0, which is the bottom right corner of the SOM in Figure 3. Figure 4 Maps of average acceleration. The distributions of mean response among the vectors in the two test data sets are presented in Figures 4(b) and 4(c), respectively. These figures exhibit similar patterns, Drug_discovery indicating that the weight vectors had converged towards the end of the SOM training. Thus, viewed in conjunction with Figure 3, it can be concluded that the SOM has learned to capture the prototype characteristics of most of the vehicle-following stimuli among the training data. The mean and variance of response associated with each neuron were next analyzed. The minimum variance of acceleration occurred at neuron (x = 0, y = 0).