Welcome to MapForums!

Register, sign in, or use Facebook Connect above to join in and participate in the forum.

When you are signed in, this message, the ads in this row, and the red-underlined link ads all go away.

Subscribe to receive our newsletter.
Subscribe Unsubscribe
Results 1 to 2 of 2

Using MapPoint for Cluster Analysis: Part 2, An Example

This is a discussion on Using MapPoint for Cluster Analysis: Part 2, An Example within the MP2K Magazine Articles forums, part of the Map Forums category; Previously we looked at cluster analysis using Microsoft MapPoint , finishing with the MPCluster add-in. Here we look at a ...

  1. #1
    Richard Marsden is offline Junior Member White Belt
    Join Date
    Jun 2011
    Posts
    5

    Using MapPoint for Cluster Analysis: Part 2, An Example

    Previously we looked at cluster analysis using Microsoft MapPoint, finishing with the MPCluster add-in. Here we look at a specific example using MPCluster.

    Again, we are using data provided by Lucassen Consulting. The objective is to find warehouse locations for a retail company operating in Western Europe. These locations will be based on delivery locations, with each warehouse located at the center of a delivery location cluster.

    Initial data is based on the number of stops by postcode. These have already been consolidated over a month, so each postcode location has multiple stops, plus total delivered weight and total number of packages. This data is available for download (link: http://dek69lxs5mxc7.cloudfront.net/...mpcluster.xlsx ) as an Excel 2007 worksheet:



    First we import this data into Microsoft MapPoint as a pushpin set using MapPoint’s Data Import Wizard. Select all data fields for display in the pushpin balloons. This is European data so we shall use MapPoint Europe 2013, but it should also work in MapPoint North America due to the presence of longitude, latitude coordinates.

    Here is the imported data with delivery locations marked with blue triangles:



    To start the clustering process, we select MPCluster: Run… from MapPoint’s Tools menu. This displays the main processing panel:




    Rather than testing all possible cluster configurations (which would take many years even for small datasets), MPCluster uses a stochastic algorithm known as “K-Means”. This starts with a “good guess” and then iteratively tries to improve it. This means that multiple runs with identical parameters can give varying results. It is also possible to apply too many constraints making stable solutions impossible (or very difficult) to find. Hence after computing a set of clusters, MPCluster will report the number of clusters found and a stability percentage. This is the percentage of data points that are stable and are not changing with each iteration. You should aim for a very high percentage – e.g. over 90%. A low percentage (e.g. 20%) typically indicates that a stable solution satisfying the constraints cannot be found. Try reducing and/or removing the constraints.

    Due to this, MPCluster can sometimes require a little bit of trial and error to find good parameters that produce a usable result.

    For this example, we only apply two constraints: the minimum and maximum number of data points (pushpins) per cluster. We shall set a cluster to have between 40 and 80 points (delivery locations).

    Some delivery locations are more important than others because they receive more deliveries. We have data for the total weight of deliveries, the number of packages, and the number of stops (ie. actual deliveries per postcode). We shall use the latter – this is the “#stops” data field. Rather than use this as a constraint, we shall use it as a weight. A constraint would add a minimum and/or maximum value for the sum of this data field. The weight will simply use the data field to pull the cluster’s center towards pushpin(s) with high “#stops”. Do this by setting the Use Data Fields and Apply a weight checkboxes, and selecting “#stops” for the weight the data field.

    Next we set the cluster display options. Clusters are marked with a boundary shape and/or a central pushpin. We shall use both. Set the shape properties by pressing the Boundary Shape button to display the Boundary Shape Colors dialog box:



    We choose a wide (4pt) red line with no fill, to maximize visibility.

    Although we do not use the ability here, MPCluster can also export all of the data points and cluster allocations to Excel. This is controlled using the Export to Excel checkbox and the Excel Options button on the main MPCluster panel.

    Finally, we choose the number of clusters to find. MPCluster does have the ability to estimate the number of clusters present in the data. This is only an estimate and should only be used after setting all of the other parameters. Press the Estimate button to display the Estimate the Number of Clusters dialog box:



    Set the range of cluster counts to search and press Start Estimation. Note that this can be time consuming for large numbers of clusters, so it is strongly recommended that you restrict the range as much as possible. For this dataset and these constraints, a range of 3 to 10 is reasonable.

    Waiting a minute or two, we come up with an estimate of ‘8’. Lets use this by pressing the Use this estimate button (this is enabled after an estimate has been calculated).

    Back on the main panel, press Start to start processing.

    Here are the results (the cluster centers have had their pushpins changed to yellow circles for visibility):



    For this run, MPCluster reported 99% stability, and has done a good job of clustering the data into groups. The Canary Islands are handled by the Spanish warehouse which is very practical. Simliarly, France is divided into three clusters: the Paris area, and two regional clusters with centers near Bordeaux and Grenoble. Italy is split into two clusters: north and south, with Sardinia and Sicily included in the southern cluster.

    From a real world perspective, the cluster centered near Bordeaux does pose a problem. First, the geometric center is in the sea – obviously we would choose a real world location on the neighbouring coast. As it is, these are only estimated locations – we would always choose actual warehouse locations near suitable infrastructure such as major highways. The Bordeaux cluster poses another problem: it includes locations in North Spain. This might not be the most economical allocation, and these locations would be candidates for re-location. It is probably more economic to supply northern Spain from the Spanish warehouse.

    MPCluster can accommodate these user adjustments. All of the shapes are standard closed freeform shapes, so it is possible to manually modify the shapes. Then, using the MPCluster Manage Clusters menu option, you may export the modified new clusters to Excel. Here is an example, with the Bay of Biscay (Bordeaux and Spanish clusters) modified as above:



    The Manage Clusters menu option also provides the ability to draw circles around the cluster centers, and to delete clusters.

    After using MPCluster to find and allocate cluster centers, sometimes it is more practical to define their territories using simple circles. The Draw Circles management function will draw a circle of specified radius around each of a cluster run’s cluster centers.

    Even with multiple colors, the map can get quite busy after multiple cluster runs. The Delete management function can be used to delete old cluster runs which are no longer needed.

    Conclusions

    This completes this example of MPCluster being used to find delivery territories and warehouse locations for a European retailer. The final analysis performed by Lucassen Consulting included multiple scenarios combined using multiple runs of MPCluster:




    Acknowledgments

    Thank you for Paul Lucassen of Lucassen Consulting for the above data.

    Further information on MPCluster including a free 14 day trial can be found at MPCluster: Cluster Analysis for Microsoft MapPoint
    Last edited by Eric Frost; 10-23-2013 at 10:15 AM.
    Winwaed Software Technology LLC
    http://www.winwaed.com
    See http://www.mapping-tools.com for MapPoint Tools

  2. #2
    huyrubiii is offline Junior Member White Belt
    Join Date
    Aug 2015
    Posts
    1

    Re: Using MapPoint for Cluster Analysis: Part 2, An Example

    thank post

Thread Information

Users Browsing this Thread

There are currently 1 users browsing this thread. (0 members and 1 guests)

Similar Threads

  1. Using MapPoint for Cluster Analysis: Part 1, Applications
    By Richard Marsden in forum MP2K Magazine Articles
    Replies: 0
    Last Post: 10-21-2013, 02:15 PM
  2. Cluster analysis
    By pinem1 in forum MapPoint Desktop Discussion
    Replies: 6
    Last Post: 07-15-2010, 01:45 PM
  3. Replies: 0
    Last Post: 06-14-2010, 09:47 AM
  4. New Product Released: MPCluster Cluster Analysis for MapPoint
    By Winwaed in forum News and Announcements
    Replies: 0
    Last Post: 04-05-2010, 11:08 AM
  5. MapBridge Census Analysis Tools for MapPoint
    By Eric Frost in forum MapPoint Desktop Discussion
    Replies: 0
    Last Post: 10-01-2002, 03:19 PM

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114