As an alternate to the statistical techniques, Computational Intelligence (CI) methods are becoming the trend for their precision in predictions, clustering, modelling and trend analysis; some methods are more popular than others and the proceeding sections presents a review of such used methods.
2.5.1
Artificial Neural Networks
Artificial Neural Networks (ANN) are now considered to be the most popular methods to deal with non-linear and ambiguous cases. For example a recent study byAmin et al.(2009) showed that ANN can act as an aggregator to multi inputs to form a single output. The author applied two ensemble methods; the negative correlation learning and bootstrap aggregating (bagging). Experimental results on a number of real-world benchmark problems showed a substantial performance improvement over other aggregator types. The authors in (Wilson et al., 2002) presented an ANN model, trained using the UK Nationwide House Price Index data to model the projected movements in property prices and to forecast future trends within the housing market. It has been shown that ANN can model any functional linear and non-linear relationship, and that such models are better than regression since regression is essentially a linear technique used to solve non-linear problems. A small scale study by (Sarlin, 2010) used Self Organising Map (SOM) to predict and monitor the financial stability, and sovereign debt for nations. It was concluded that SOM is considered to be a feasible tool for aggregating multiple related variables to visualize and monitor the evolution of economic conditions over time.
2.5.2
Fuzzy Logic
Two related studies carried out by Ammar et al. (2004); Shnaider and Haruvy
2. Literature Review
weighting and aggregating. For example, (Shnaider and Haruvy,2008) compared the performance of fuzzy logic with linear regression as modelling tools for deter- mining the constituent factors in the assessment of economic growth of national economies. Also,Keller(2008) introduced a multi-input, single-output fuzzy con- troller to act as an artificial decision maker that operates in a closed-loop system and in real time to forecast and control the dynamics of macroeconomic variables using a fuzzy learning algorithm. It was found that the fuzzy logic yielded more stable and consistent results than those of linear regression.
A few other attempts were aimed at creating fuzzy indicators that use “pure” fuzzy logic to evaluate and rank nations sustainability performance, for exam- ple, the sustainability assessment by fuzzy evaluation (Andriantiatsaholiniaina et al., 2004;Phillis et al.,2011) introduced “experts” based fuzzy rules that uses a fuzzy weighted sum of inputs which was computed and assigned to the output variable. Abouelnaga et al.(2010) offered a Nuclear Energy Sustainability Index, using fuzzy logic, which they based it on the same methodology offered by An- driantiatsaholiniaina et al. (2004). Yet again, experts’ rule based systems suffer from generic problems such as the subjectivity and possible biases of the experts who devise the rules; the credibility and the hidden identity of who the authors usually call “experts”, also the overlapped variables which mostly would generate unmanageable size and tangled rules.
In regards to treating missing data, Hathaway and Bezdek (2001) proposed four new techniques which can be integrated with Fuzzy c-Means (FCM) to allow it to accept and cluster incomplete datasets. Another study by Nuovo (2011) have applied such the aforementioned technique to show that one can indeed use FCM derived strategies, to precisely impute missing data.
2.5.3
Hybrid Techniques
Studies by Kershaw and Rossini (1999) and Dost´al (2009) incorporated hybrid methods like fuzzy logic, traditional econometric techniques, and genetic algo- rithms to develop constant price index and stock market decision machine. The work indicated that such methods could be integrated to present a real alterna- tive to the econometric methods or to improve prediction accuracy. Another inte-
2. Literature Review
grated methods by Liu(2007) used Multiple-Criteria Decision-Making (MCDM) technique and fuzzy logic to calculate environmental sustainability, rank and clus- ter nations. The framework considered five components: air, water quality, water quantity, land use and natural resources. Studying such methods, one would ex- pect that the authors had figured a way to extract rules using the MCDM to feed into the fuzzy system to create a coherent and intelligent rules to govern the fuzzy part of the framework. However, the use of MCDM did not serve in creating the rules for the evaluation by the fuzzy logic, instead the framework consisted of two separate techniques that do not complement each other. To set the weights and create the rules the authors resorted to the classic “IF-THEN” fuzzy rule based system which was put by a panel of three “experts” to govern the framework. As stated before, experts rule based systems suffer from common problems such as the partiality and subjectivity of the experts who create the rules. The trust- worthiness and the concealed identity of the “experts”, in addition to producing large size and entangled rules.
Regularly, the Arithmetic Mean (AM), Geometric Mean (GM), Additive Rules (AR) etc. are used for the purpose of aggregating the variables to form a single value, hence a “composite index” (OECD, 2008a). On other hand, computa- tional intelligence techniques, such as ANN, SOM and fuzzy rule-based systems and recently some hybrid methods such as hesitant fuzzy geometric means, and intuitionistic fuzzy hybrid geometric operators have been proposed and applied to act as an aggregator for multi input-single output systems. Such methods were made to help decision makers to effectively deal with multiple attribute decision making under hesitant or intuitionistic environments (Zhu et al., 2012;Zhao and Wei,2013).
Panel data regressions or Time-Series Cross-Section (TSCS) regressions mixed with some CI methods has been astonishingly neglected, but a unique study by
Pao and Chih (2006) concluded that ANN models can be used to solve panel data regression, and that it would allow to construct and test sophisticated mod- els than purely cross-sectional or time-series data to solve debt policy forecasting problems. Saisana and Munda (2008) proposed using sensitivity analysis when deciding on what to measure, and what to include/exclude from different indi- vidual indicators. Herrero et al. (2011) incorporated three different methods,
2. Literature Review
including Cooperative Maximum Likelihood Hebbian Learning, SOM and Curvi- linear Component Analysis to select the most appropriate variables to forecast the political risk for most of the world’s nations. However, the suggested tech- niques are either heuristics or too technical which do not measure the reality of the developments pressing issues.