FIR–based Classifiers for Animal Behavior Classification Majid M. Beigi and Andreas Zell Abstract—In this paper, we implement a new method for classification of biological signals in general, and use it in the animal behavior classification as an example. The forced swimming test of rats or mice is a frequently used behavioral test to evaluate the efficacy of drugs in rats or mice. Frequently used features for that evaluation are obtained through observing three states: immobility, struggling/climbing and swimming in activity profiles. We consider that those activity profiles (signals) inherently contain undesired and interference noise that should be removed before feature extraction and classification. We use a Finite Impulse Response (FIR) filter to filter out that additive noise from the activity profile. The parameters of the FIR filter are obtained via maximizing the accuracy of a classifier that tries to make a discrimination between two classes of the activity profiles (e.g. drug vs. control). We use the kernel Fisher discriminant criterion as a criterion for the discrimination, the DIviding RECTangles (DIRECT) search method for solving the optimization problem and Support Vector Machines (SVMs) for the classification task. We show that Autoregressive (AR) coefficients are suitable features for the extraction of the dynamic behavior of rats and also the classification of activity profiles. Our proposed behavior classi- fication method provides a reliable discrimination of different classes of antidepressant drugs (imipramine and desipramine) administered to rats versus a vehicle-treated group. I. I NTRODUCTION The Forced Swimming Test (FST) is a behavioral test used frequently to evaluate the potential efficacy of drugs affecting the central nervous system (CNS) in rats or mice [3]. In this experiment, rats are exposed to a 15-min pretest swim period and followed the next day by a 5-min test swim. Immersion of rodents in water for an extended period of time produces a characteristic behavior called immobility, in which the rat makes only those movements necessary to keep its head above water. When antidepressant drugs are administered between the pretest and test periods, usu- ally three times within 24hr, the behavioral immobility is selectively decreased [4]. Depending on the type of drug, rats show a mixed behavior of activities such as immobility, struggling/climbing (the rat tries to escape from the water) and swimming. Researchers have tried to conclude the effect of drugs from the above three states (immobile, struggling and swimming) [25]. Typically, tricyclic antidepressants and drugs with selective effects on noradrenergic transmission increase struggling/climbing behavior, while selective sero- tonin reuptake inhibitors increase swimming behavior versus the control group [7], [5], [6]. M. M. Beigi is PhD. student at the Department of Com- puter Science, University of Tuebingen, 72076 Tuebingen, Germany. majid.beigi@uni-tuebingen.de A. Zell is Professor at the Department of Computer Science, University of Tuebingen, 72076 Tuebingen, Germany. Andreas.Zell@uni-tuebingen.de (a) Tylose (b) Desipramine 30 mg Fig. 1. FST test as a behavioral test. The successive images of rat movement are converted to activity profile signals. a) activity profile (arbitrary unit) for tylose as control and b) Desipramine 30mg as antidepressant. Fig. 1 shows examples of activity profiles, which are gained from successive images of rat movement in FST test. Considering predefined thresholds for the immobility, strug- gling and swimming states, and depending on the amplitude of the activity profile, the average period of time in each state can be measured. By comparing those parameters for the animals treated with an antidepressant against the control group, which was treated with the vehicle, it can be observed, for example, whether the swimming behavior of rats with an antidepressant drug has increased. In an automated classification method, we aim to classify animals treated with known antidepressants and the control group. However, our experiments show that the response of the rat to drugs is too complex to only consider those states as indicator of the drug efficacy. The detection of the behavior of rats depends on recognizing changes in some characteristics of movement and we are interested in features which represent also the dynamic behavior of rats. An important issue in the automated classification is the presence of noise in activity profiles. We consider the activity profile, x(n), as the total of the inherent response of a rat to a drug at a certain dose, s(n), and the undesired and