Lazaros S. Iliadis

A DECISION SUPPORT SYSTEM APPLYING AN INTEGRATED FUZZY MODEL FOR LONG- TERM FOREST FIRE RISK ESTIMATION

Abstract Fire is the main cause of forest destruction in Mediterranean basin countries. Long-term prediction is intended for long-term planning which may serve to characterize and cluster regions as subject to high or low fire risk. This will enable the development of a rational and sensible forest fire prevention and protection policy. The problem with the existing approaches of long-term forest fire risk clustering is that they use crisp sets applying specific cluster boundaries. On the other hand, fuzzy algebra can provide reliable and flexible means of modeling that can be applied by a suitable decision support system. Given a specific area of interest, the evaluation of the long-term forest fire risk can be performed by the use of a triangular and a trapezoidal membership function. The decision support system that has been developed applies an inference mechanism that is based on various aspects of fuzzy sets and fuzzy machine learning techniques. The system has been applied in Greece, but it can be used on a global basis. Results show that the system successfully estimates the forest fire risky areas.

An Artificial Neural Network model for mountainous water-resources management: The case of Cyprus mountainous watersheds

This is a preliminary attempt towards a wider use of Artificial Neural Networks in the management of mountainous water supplies. It proposes a model to be used effectively in the estimation of the average annual water supply, in each mountainous watershed of Cyprus. This is really a crucial task, especially during the long dry summer months of the island. On the other hand the evaluation of the potential torrential risk due to high volume of water flow in the winter season is also very important. Data (from 1965-1993) from 78 measuring stations located in the 70 distinct watersheds of Cyprus were used. This data volume was divided in the training subset comprising of 60 cases and in the testing subset containing 18 cases. The input parameters are the area of the watershed, the average annual and the average monthly rain-height, the altitude and the slope in the location of the measuring station. Consequently three structural and two dynamic factors are considered. After several and extended training-testing efforts a Modular Artificial Neural Network was determined to be the optimal one.

A computer-system that classifies the prefectures of Greece in forest fire risk zones using fuzzy sets

Abstract All of the Mediterranean countries face a serious forest fire problem. The main factors that affect the problem of forest fires in Greece are vegetation, climate conditions and most of all, arson (Proceedings of Forest Fires in Greece, Thessaloniki, 1990, p. 97). In Greece, after 1974, the number of forest fires and the total burned areas have risen dramatically. The design of an effective fight and prevention policy is a very important matter, as it can minimize the destruction. This paper describes an expert system that classifies the prefectures of Greece into forest fire risk zones, using a completely new methodology. The concept of fuzzy expected intervals (F.E.I.) was defined by Kandel and Byatt (Proc. IEEE, 66, 1978, 1619) and offered a very good approach towards forest fire risk classification. Fuzzy expected intervals are narrow intervals of values that best describe the forest fire problem in the country or a part of the country for a certain time period. Fuzzy logic was applied to produce a F.E.I. for each prefecture of the country. A successful classification of the prefectures of Greece (in forest fire risk zones) was performed by the expert system by comparing the produced fuzzy expected intervals to each other and by using a supervised machine learning algorithm that assigns a certain weight of forest fire risk to each prefecture (Machine Learning, John Wiley and Sons, 1995).

Wood-water sorption isotherm prediction with artificial neural networks: A preliminary study

Abstract This is a preliminary study that proposes an original prototype artificial neural network to be used in addition to the two classic sorption isotherm modeling methods, Hailwood-Horrobin (HH) and Guggenheim-Anderson-deBoer (GAB), in predicting the equilibrium moisture content in wood at three different temperatures (30, 45 and 60°C) for softwood (lodgepole pine) sapwood and heartwood specimens. Contrary to the HH and GAB equations, which use physical data for modeling, the predictive power of the artificial neural network is based on both physical and chemical data for the specific wood types. The results prove the potential efficient use of neural networks in predicting moisture content based not only on the ambient conditions, but also on taking into consideration the chemical composition of wood.

Neural network prediction of bending strength and stiffness in western hemlock (Tsuga heterophylla Raf.)

Abstract The stiffness and strength, modulus of elasticity (MOE) and modulus of rupture (MOR), as well as density, moisture content, microfibril angle and diffraction pattern coefficient of variation of azimuthal intensity profile (ICV) was determined for 259 small clear specimens. These samples represent 38 old- and second-growth western hemlock (Tsuga heterophylla) trees harvested from several sites in coastal British Columbia, Canada. The data were analyzed by classic statistical regression techniques to reveal interrelations among the mechanical properties and the inherent wood properties. Simultaneously, the predictive power of artificial neural networks was evaluated with the same data set by employing several optimization techniques. Regression analysis of wood density and the flexural strength properties resulted in R2 of 0.172 and 0.332 for MOE and MOR, respectively. The most efficient network model proved to be far superior demonstrating correlation coefficients with models for MOE ranging between 0.693 and 0.750, and the corresponding MOR models ranging between 0.438 and 0.561 in all testing phases. It is apparent that neural networks have the potential and capacity to self-train and become powerful adaptive systems that can predict the strength and stiffness of wood samples. The neural network analysis also revealed the importance level of each independent variable on both MOE and MOR properties.

Wood dielectric loss factor prediction with artificial neural networks

An artificial neural network that can predict the dielectric properties of wood was developed and tested with experimental data. The network was capable of accurately predicting the loss factor of two wood species not only as a function of ambient electro-thermal conditions but also as a function of basic wood chemistry. This way, an important predictive tool is created that allows optimization of dielectric heating and drying for many wood species without significant experimentation should their chemical composition be known under variable temperatures, moisture contents and electric filed characteristics.

Fast and low cost prediction of extreme air pollution values with hybrid unsupervised learning

Air pollution is the problem of adding harmful substances or other agents into the atmosphere and it is caused by industrial, transport or household activities. It is one of the most serious problems of our times and the determination of the conditions under which we have extreme pollutants’ values is a crucial challenge for the modern scientific community. The innovative and effective hybrid algorithm designed and employed in this research effort is entitled Easy Hybrid Forecasting (EHF). The main advantage of the EHF is that each forecasting does not require measurements from sensors, other hardware devices or data that require the use of expensive software. This was done intentionally because the motivation for this work was the development of a hybrid application that can be downloaded for free and used easily by everyday common people with no additional financial cost, running in devices like smart phones. From this point of view it does not require data from sensors or specialized software and it can offer people reliable information about extreme cases.

On-line consistent ranking on e-recruitment: seeking the truth behind a well-formed CV

In this work we present a novel approach for evaluating job applicants in online recruitment systems, using machine learning algorithms to solve the candidate ranking problem and performing semantic matching techniques. An application of our approach is implemented in the form of a prototype system, whose functionality is showcased and evaluated in a real-world recruitment scenario. The proposed system extracts a set of objective criteria from the applicants’ LinkedIn profile, and compares them semantically to the job’s prerequisites. It also infers their personality characteristics using linguistic analysis on their blog posts. Our system was found to perform consistently compared to human recruiters, thus it can be trusted for the automation of applicant ranking and personality mining.

Fundamental fuzzy relation concepts of a D.S.S. for the estimation of natural disasters' risk (The case of a trapezoidal membership function)

The effective protection from natural disasters requires the development of a rational and sensible protection and prevention policy. This project deals with the development and testing of a decision support system that acts on two levels. On the first level, it estimates the annual forest fire risk for each area of Greece using a fuzzy Trapezoidal membership function. Reference is done to past work in this area and past results of forest fire risk estimation (using other models) were compared to the results of this system. On the second level, it forecasts a narrow expected closed interval for the burned area, using a fuzzy expected interval model. It is the first time that such forecasts are produced and such results are obtained. Physically and operationally, the decision support system consists of two parts. The risk estimation part is more straightforward and it was developed in MS-Access in order to have the ability to store and use a vast amount of data. The forecasting part was developed in a decision support system shell, in order to have a sound Inference mechanism.

Application of fuzzy T-norms towards a new Artificial Neural Networks' evaluation framework: A case from wood industry

The development of an Artificial Neural Network requires proper learning and testing procedures that adopt error correction processes and algorithms. Monitoring of processing elements values and overall performance is one of the most critical issues of an Artificial Neural Network development process. This should happen as the network evolves and it is the actual task that enables the developer to make informed decisions about the proper network topology, math functions, training times and learning parameters. This manuscript presents an innovative and flexible error validation framework applying fuzzy logic. It offers an approach capable of viewing the task of performance improvement under several different perspectives. Then the developer has the capacity to decide which performance is most suitable according to his standards. The model has been tested for a specific industrial case study with actual data and a comparison to the existing methods is presented.