engleski

Towards a High Performance Expert System for Gait Analysis

In a number of decision support systems searching through large data warehouse based on sample sequence is desirable. Gait analysis systems store demographic data (categorical, numerical data) and time sequence data (human gait variables) where query by example is needed for matching signal patterns. Although several different approaches have appeared, most of them are not specialized for the problem which combines simple facts and time series data or data warehousing principles for such data. We propose an end-to-end approach including data warehouse partitioning, symbolic-based algorithms, clustering-based algorithms for purpose of fast and intelligent retrieval and classification of stored facts. This paper gives the modified version of symbolic algorithm that allows dimensionality reduction and indexing with a lower-bounding distance measure. Paper shows how it can be used together with clustering based algorithms for efficient processing of timeseries data stored in large fact tables of a data warehouse. Each symbolic word serves as input to hierarchical clustering algorithm and forms a node in the dendrogram representation (graph which displays nodes arranged into hierarchy). In a data warehouse this design improves performance in addition to denormalization, by adding derived information and summary tables. Input signal is divided into segments to reduce the length of the symbolic words. Since each word is represented as a simple fact in a database, hierarchical clustering algorithms can be applied. We have tested this approach on human locomotion data combining categorical, numerical and time series variables. Input dataset included ground reaction force during a simple step, representing a subset of standard gait variables, demographic and anthropometric parameters. The measurements were performed in the Biomechanics Laboratory at the Faculty of Kinesiology, University of Zagreb, using 2002-ELITE system with Kistler platform (40 cm by 60 cm).

human gait; data warehouse; ground reaction force; dimensionality reduction; time series

nije evidentirano