Comparison of three different approaches to the property prediction problem

Neighborhood subspace approximation method has been developed for solving the property prediction problem. In this paper the performance of this method is evaluated on three groups of compounds. The molecular structure was encoded as sequences of well-known structural (topological) indices. Euclidean distance has been used for determining the similarities between compounds. Property prediction results of the new method are compared with the results of the neighborhood based approximation (clustering) method and with the results obtained by an artificial intelligence program using machine learning tools.