A variety of computerized systems, such as tracking systems producing X,Y coordinates to classify behavior (e.g. EthoVision [1]) and coding systems using operationally defined behavioral items (e.g. The Observer [3]), has been developed for the analysis of behavior of animals and man.

One has attempted to improve the quality of tracking systems by enhancing the spatial resolution and the sampling rate. However, these improvements do not solve the problem of "splitting" behavioral items, a traditional issue in ethology. For instance, it remains a matter of definition to describe "walking" and "non-walking" as "moving over a minimum distance of 10 cm/s" and "moving less than 10 cm/s", respectively. This problem is the consequence of assessing classic logic in which crisp sets such as the above-mentioned Yes/No splitting are used.

The mentioned problem can be solved by assessing fuzzy logic, a multivalued logic, that allows the merging of data collected with tracking systems and data collected with coding systems. As a result, an infinitive number of alternatives between Yes and No notations can be mathematically formulated and processed by computers. In this way it becomes possible to differentiate behavioral items such as walking at normal speed and walking just slower than normal on the basis of the calculated confidence factors.

We will present a method in which fuzzy logic is used to classify behavior of freely moving rats that has been recorded on videotape. The first step of this method implies the classification of postures of rats with the help of Fourier descriptors and a neural network. This results in a coding system in which four-digit codes are used to label postures [2]. The second step of this method implies the calculation of the position (X,Y) and the orientation of the rat in the open field with the help of a tracking system. Finally, fuzzy logic is used to merge both sets of data in order to classify behavior.
We will illustrate that this approach helps to solve the problem of the subjective attribution of labels to behavioral elements as it is normally done in coding systems using operationally defined behavioral items. Apart from this, we will illustrate that this approach can also be used for attributing a mathematically calculated confidence factor to each classified behavioral item.

Figure 1. An example of a coding system, in which a four-digit code has been assessed to assign distinct postures.

The method including the software program has originally been developed for the analysis of behavior. However, it is generally applicable in a great variety of research areas.

References

1. Buma, M.; Smit, J.; Noldus, L.P.J.J. (1997). Automation of behavioral tests using digital imaging. Neurobiology, 4, 277.
2. Heeren, D.J.; Cools, A.R. (2000). Classifying postures of freely moving rodents with the help of Fourier descriptors and a neural network. Behavior Research Methods, Instruments & Computers, 32, 56-62.
3. Noldus, L.P.J.J.; Trienes, R.J.H.; Hendriksen, A.H.M.; Jansen, H.; Jansen, R.G. (2000). The Observer Video-Pro: new software for the collection, management and presentation of time-structured data from videotapes and digital media files. Behavior Research Methods, Instruments & Computers, 32, 197-206.

Paper presented at Measuring Behavior 2000, 3^rd International Conference on Methods and Techniques in Behavioral Research, 15-18 August 2000, Nijmegen, The Netherlands

© 2000 Noldus Information Technology b.v.