Meta-Monitoring Using an Adaptive Agent-Based System one person; thus, there is a duplication for each individual looked after. None of these systems collects individual monitoring for merging global behaviour patterns (Rammal, Trouilhet, Singer, & Pécatte, 2008). Nevertheless, patterns of moni- tored people could be used to estimate the status of someone in relation to their community or to integrate new comings. To collect numerous individual patterns and build collective one, our system must be able to handle complex data. We can't rely on a central- ized algorithm because data is heavily distributed, not always available and often partial. First, we can't imagine a single system to monitor every person in his home. Second, each person will be equipped with his own set of apparatus, leading to heterogeneous data. Classical algorithms are inadequate in such a context and only approxi- mate, fault tolerant and flexible solutions can be proposed. · · · each classification agent selects the subset of data to classify among the data available; each classification agent chooses the most suitable classification algorithm; each classification agent computes the most adequate settings to handle its data set (normalization and weight of attributes). Let O 1 , ...,O n be objects to classify, each object is described by p numerical attributes X 1 , ..., X p , so each object O i is represented by a vector i i ( x 1 , ... , x i j , ... , x p ) where x i j is the value of attribute number j of O i . x i j can be normalized by two ways: Normalization between [0;1]: i x i j - x min j