A Pattern Mining Approach for Classifying Multivariate Temporal Data

Abstract

We study the problem of learning classification models from complex multivariate temporal data encountered in electronic health record systems. The challenge is to define a good set of features that are able to represent well the temporal aspect of the data. Our method relies on temporal abstractions and temporal pattern mining to extract the classification features. Temporal pattern mining usually returns a large number of temporal patterns, most of which may be irrelevant to the classification task. To address this problem, we present the minimal predictive temporal patterns framework to generate a small set of predictive and non-spurious patterns. We apply our approach to the real-world clinical task of predicting patients who are at risk of developing heparin induced thrombocytopenia. The results demonstrate the benefit of our approach in learning accurate classifiers, which is a key step for developing intelligent clinical monitoring systems.

Figure 1

An example illustrating the trend and value abstractions. The dashed lines represent the 25th and 75th percentiles and the solid lines represent the 5th and 95th percentiles.

Figure 2

A temporal pattern with states 〈A₁, B₂, C₃, A₂〉 and temporal relations R_1,2 = c, R_1,3 = b, R_1,4 = b, R_2,3 = c, R_2,4 = b and R_3,4 = c.

Figure 3

A high-level description of candidate generation. The algorithm takes as input the frequent k-patterns (F_k) and returns the candidate (k+1)-patterns (Cand) together with their pid-lists.

Figure 4

The number of patterns return by TP_all, TP_chi, MPTP and A-MPTP for different minimum supports.

Figure 5

The running time of TP_Apriori, TP_id-lists, MPTP and A-MPTP for different minimum supports.