Segmentation of Continuous Data Streams Based on a Change Detection Methodology - Advanced Techniques in Knowledge Discovery and Data Mining - page 117

4.3.3 Results - “Stock” Data Set

This section summarizes segmentation and analysis of the stock data set, which
was analyzed by the IFN algorithm. The main expectation for this dataset was that
significant changes would be observed over time, due to the segmentation of the
full data sets into disjoint segments. This indicates that the full data stream can be
evaluated as several disjoint data sets, and for each of them, a separate underlying
model can be evaluated and implemented.
The full data set holds information about stocks in 5462 records. As there is no
predefined way to segment the given data stream, three different ways of
segmentation were implemented and evaluated based on the change-detection
methodology.
The logical way of segmenting the data stream is using any “time field” as an
indication of the accumulating knowledge, which was added incrementally to the
database. Table 4.6 describes how the segments of data sets are divided according
to the incremental date field in the stock data set.

Table 4.5. Segmentation of the “stock” data set.
Trial num. Segment num. Record interval

1 [1,1000]

2 [1001,2000]

1 3 [2001,3000]

4 [3001,4000]

5 [4001,5000]

1 [1,1500]

2 [1501,3000]

2

3 [3001,4500]

4 [4501,5000]

1 [1,2000]

2 [2001,2500]

3 [2501,3000]

3 4 [3001,3500]

5 [3501,4000]

6 [4001,4500]

7 [4501,5000]

The first trial is a partition of 5000 accumulated records into five equally sized
data sets. Figure 4.3 describes the outcome of applying the change-detection
methodology to these segments of data.