We should try to choose parameters that are robust to noise. In the above example, the start value and the end value are both sensitive to noise, as compared to, say average and gradient, both of which are aggregate measures of the data points belonging to the period of increase. Hence it would be better to represent an increasing event using the latter two parameters rather than the former. In general, an aggregate of the signal will be more robust than a single measurement of the signal, so average value and gradient are robust than start value and end value.