Last month we hosted a unique GeoDev Webinar when we had Manushi Majumdar share her presentation on "Thinking Spatially and Statistically". Manushi introduced types and characteristics of spatial data and advanced GIS analysis techniques. She was able to cover a few basic concepts of statistics and show how they differ in a spatial context, advancing towards Spatial Machine Learning with ArcGIS.
Here are the questions that were received during the webinar along with their respective answers:
Q: What is the difference between machine learning and statistics? For example, with regression, is there a difference? This always puzzles me!
A: Here is a resource to understand the difference: https://www.kdnuggets.com/2016/11/machine-learning-vs-statistics.html
Q: Is there any geo-processing tools in built in ArcGIS for running machine learning algorithms?
A: Yes, ArcGIS has support for several Machine Learning techniques. We would suggest looking this blog post to learn more about ML support within ArcGIS Desktop. https://www.esri.com/arcgis-blog/products/arcgis-pro/analytics/machine-learning-in-arcgis/
Q: Can you provide the link to the notebook again?
Q: Are there any other good resources for finding examples of utilizing Machine Learning with GIS?
A: Here are just a few ArcGIS blogs demonstrating examples:
Q: Spatial Join: I see the tool has capability to join two layers with out common attributes . But can this be done on multiple layers in a single shot? The built-in tool has only option to select two layers. What are the options?
A: Join works on a 1:1 principle, you can only join one layer to another. That said you can use the concept of table 'Relate' to join one table to many using a common attribute in those tables (does not work spatially).
Q: Is it possible to integrate ArcGIS with machine learning software like Jupiter notebook?
A: You can use ArcPy as well as the ArcGIS API for Python in Jupyter notebooks.
Q: Is there a way to use machine learning to predict or project possible future incident locations without assigning a z-value?
A: Z-score (standard score) denotes the number of standard deviations from the mean a data point is. Simply put, it conveys the distribution of a point around the mean. Prediction or Classification does not need z-scores for input variables.
Q: Could you walk us through the hotspot analysis? How do you access these tools?
A: Read through this https://pro.arcgis.com/en/pro-app/tool-reference/spatial-statistics/h-how-hot-spot-analysis-getis-ord-gi-spatial-stati.htm to learn more about HotSpot Analysis. It can be accessed within the Spatial Statistics (Mapping Clusters) toolbox in ArcGIS Desktop and under the Analyze Patterns category in ArcGIS Online.
Q: Can we do the machine learning analyses using 10.6 geoprocessing tools?
A: Yes, apart from the usual, ArcGIS Desktop 10.6 comes with two new tools Deep Learning Model To End and Export Training Data For Deep Learning.
Q: I have a GIS online account. How can I access the data demonstration in ArcGIS online?
A: The data I used for my examples is publicly available. Once you add it to your ArcGIS Online account, you can use the Summarize Center and Dispersion tool there to generate spatial mean, median and standard dispersion for your data.
Q: How can I use ArcGIS for linear regression or logistical regression analysis?
A: Support for regression, both linear and logistic, is available in ArcGIS Desktop Spatial Statistics (Modeling Spatial Relationships) toolbox. Click here https://pro.arcgis.com/en/pro-app/tool-reference/spatial-statistics/an-overview-of-the-modeling-spatial-relationships-toolset.htm to learn more.
Q: Is machine learning part of programming, or is it remote sensing?
A: Machine Learning involves concepts of statistics as well as algorithms to solve problems based on patterns or inferences drawn from data. Remote sensing, on the other hand, involves studying the planet using remote instruments. Machine Learning can have applications in the field of Remote Sensing, for instance, to detect buildings, roads using satellite imagery data.
Q: Is the Jupyter environment embedded directly within Esri ML module?
A: ArcGIS Enterprise 10.7 comes with Hosted Notebooks, that lets you perform spatial analysis and data science workflows in a notebook within your portal. Other than that, you can use ArcPy or ArcGIS API for Python in an external Jupyter Notebook too.
Q: Which interpolation techniques suits best when you are dealing with underground water data?
A: While it depends on your sampling size and distance, Kriging might be a good Interpolation technique.
Q: Please suggest out-of-the-box tools provided by ArcGIS for machine learning algo.
A: Yes, ArcGIS has support for several Machine Learning techniques. I'd suggest looking this blog post to learn more about ML support within ArcGIS Desktop. https://www.esri.com/arcgis-blog/products/arcgis-pro/analytics/machine-learning-in-arcgis/
For more information, Manushi shared her presentation: GeoDev Webinar - Thinking Spatially and Statistically
Also, for the full recording of the webinar, click here.