ABSTRACT
According to Greek historian Herodotus, around 3050 BC, the ancient Egyptian pharaohs 
collected data about country's wealth and population in order to prepare for construction of 
the pyramids. This effort to compile great and diverse amounts of information just might 
correspond to the historical beginning of statistics. Additionally, the use of advanced 
techniques in data management known as data mining was coined in the early 1990s, based 
on fields of science and knowledge such: databases; information retrieval; classical 
statistics; machine learning; decision making process; data visualization; parallel and 
distributed computing; and others such as, natural language; image analysis; signal 
processing; computer graphics, etc.
The general objective of this PhD thesis is the treatment and handling of data applied to 
Water Supply Systems (WSS) from the paradigm Knowledge Discovery in Databases 
(KDD), specifically using data mining methods. Due to the variety of problems that could 
be solved from the information obtained during design, operation and management of WSS, 
we consider that this is an appropriate field for a KDD application model. Many other 
branches of science and knowledge have been explored by the KDD model. Although water 
is a natural renewable resource, it has limitations with respect to quantity and quality for 
each location and specific time. Engineering's scientific community and WSS managers 
have shown a growing interest in the development of innovative data management 
techniques, but much remains to be learned. The KDD and data mining techniques can help 
us to support the resolution of questions to be developed, managed and corrected in WSS.
We developed this PhD thesis to contribute to this effort, and we have taken rigorous care 
to ensure that proposals and discussions presented here enable us to understand the use of 
KDD in the management of WSS. Specifically, we used the available information and the 
selected tools to generate a management model based on decision rules to deal with 
reported damages in the water network.
The first step of the methodology was to conduct a comprehensive theoretical framework 
study of KDD. Since most investigations and developments of this subjects have been 
published in English we present it in Spanish. Next, we developed an exhaustive state-of-
the-art study for the investigations involving the use, application and development of KDD 
methodology in WSS, and we describe it in detail. 
Next, we present a practical application to find the damage that occurred during 2006 in the 
water supply network of Calarcá, a town located in the Colombian coffee region. (The 
water supply company reported these damages.) To develop the application, we searched 
relationships between variables found. Data used included information reports, the 
hydraulic model and risk factor level maps that measure natural disasters affecting the 
Colombian coffee region. Some of the information used for this application was given to us 
by the public-private partnership company Multipropósito of Calarcá S.A. ESP, which 
manages the water supply in the town.
After KDD's steps were done and appropriate models were chosen, we decided to use the 
following data mining tools: classification and regression trees, neural networks and 
Kohonen networks. The data mining software used was SPSS Clementine 9.0. The best 
results were obtained with the classification and regression algorithms, though these results 
do not indicate a strong dependency between variables. Nevertheless, with the outcomes of 
modeling and the future developments proposed, we can obtain a framework based on the 
information generated by the system itself. This tool would assist us in the management and 
solution of problems regarding the design, operation and management of a WSS.
The results obtained, besides their great potential for WSS applicability, can be improved 
with basic information taken for this purpose, as reflected in the final recommendations.
The methodology and the studied model have the advantage of continuous updating as well 
as real-time decision-making capability which undoubtedly provide a powerful and useful 
management tool. Most interestingly, the tool comes from the actual information of the 
supply network itself, which eliminates the problem of uncertainties that arise when 
modeling the network, as well as subjective assessment of parameters included in these 
formulations.
Finally, we propose some future courses of action to improve the available information and 
the research to achieve better results than those obtained up to this point.