The return from Bolivia was tiring, 22 hours of travel and the most complicated thing was being on the last stopover stuck in the airport of Comalapa, El Salvador before arriving in my starting country. It was a tiring week, 8 to 5 work days sitting most of the day, lots of food, but also a lot of learning.
Almost all of us have concluded that the course has been too loaded with content and very little practical work, this affects the burden on an instructor who must handle an entire day's presentation, with half-boring Powerpoints and an audience of different levels ... half doze the other half lost and a few looking for a practical benefit to what they already do. However, the CD with the presentations and the complement with exhibitions from various countries has brought good results.
Among the presentations, the one that has caught my attention the most is the application of neural networks to complex processes under the principle of artificial intelligence.
Whether it is done by a central institution or a local municipality, to collect property tax requires implementing a massive valuation methodology. To do this there are several from simplified (liars) to too complex (unsustainable). One of these widely disseminated methodologies is through the market method for the valuation of land and replacement cost for buildings. This requires at least three strenuous tasks:
1 Update of values of improvements. Its instrumentation is through what is known as constructive typologies, these are built with budgetary chapters, which in turn are made up of constructive elements and composed of basic ones as unit cost sheets. In such a way that the simplest thing is to update the input base: materials, labor, equipment and machinery, more professional services and then the construction typologies are ready to be applied. The practicality of methodologies like this one is that the collection of field data for the valuation form only requires calculating the construction area, construction characteristics, quality and conservation ... well documented, it can overcome subjectivity.
For rural areas, a study is also made of those characteristics that give the property a productive value, such as permanent crops, tradable resources or potential use.
2 Map update of earth values. This is built based on a sample of reliable real estate transactions, with a significant representation and projected in time to have the market value. Then these values become homogeneous zones that contain a trend based on proximity and services.
3 Network Update public services. It happens that when the state of road infrastructure changes, to give an example, these characteristics affect a property on one or more of its fronts. Therefore, it is ideal that the values are transferred from the block to the street axis so that they can be associated with the proportion that affects the front of the property ... ideally, that the area has certain characteristics that give it a value for service networks and relationship of neighborhoods to benefits that affect not only the value of the land that can be very linear.
Doing it every 5 years is not difficult, but doing it in a differentiated way for many municipalities becomes unsustainable madness even if there is a computer application, because it still depends on external data and field samples.
Yedra García, from the Ministry of Economy of Spain has presented a presentation on the subject "Artificial intelligence applied to mass valuation"
The concept is there on the web, in English, however Yedra has raised a possibility, through the use of neural networks that applied to this problem would solve the automation of the methodology as complex as it may seem:
It means that a minimum number of indicators at medium level, can have a comparative relationship that by sending down a trend of input values and upwards a tentative proposal of values of homogeneous areas through spatial analysis by similarity of conditions, can generate a matrix which makes redundancy in both ways against real data, such as data from electronic bulletins of construction prices or real estate values.
Of course, this does not lead to a simple analysis of tabular data, but also a spatial analysis of layers that affect valorization, interconnection of road trunks and topological analysis of shared neighborhood.
This could bring results beyond the simple valuation for property tax purposes, such as the planning or planning of works based on the conditions of impact on the revaluation and recovery of capital gains ... among others.
The posture leaves me the green smoking itch someday in the intention of implementing it.