Figure 1.  Steps of the KDD[3] Process.

The appearance of these rules has started with the use of large databases in commercial transactions known as Big Data [19] [20]. This is more generally known today as the classic problem of “market-basket of the housewife” which explains the origin of rules associations. This is perhaps best illustrated by the relation˝ if condition then result˝. The question then is to determine for each basket purchased by a customer, expressing his needs and preferences, the rules on which the supermarket should focus to manage efficiently all the customer baskets simultaneously. It will then be necessary to coordinate very important mass of information to be able to define, analyze and extract relevant information for that target.

In our case it will be necessary to associate individual characteristics of the unemployed to determine duration of unemployment related to them, and at the same time to be able to react on pre-established rules decision by selecting actions to reduce it, and consequently the probability of outflows of unemployment. Using KDD[4] techniques, for such massive detailed individual database, translated into relevant and synthetic information using “Data Mining procedure”[5]should make us then reach our target.

The result will then be expressed by associative algorithms and will eventually give simplified computer model for very complex rules associations. Data mining technique generally include different associations and fields from self-learning, statistics, knowledge representation, artificial intelligence, expert systems. It is an iterative and interactive analysis scheme which uses raw data to extract relevant and easily proceedable and reliable information by the analyst [21].The interactive process generated by Data Mining reflects the way that human analyst Could analyse, control, and take corrective decisions. The analyst takes the charge of reprocessing the information to extract the most interesting. This must be preceded by a pre-selection that makes search on data easier. Then the procedure will consist on applying artificial intelligence methods to determine the best algorithmic model. At this level, various graphic and numerical visualization methods will enable the various results to be presented and evaluated effectively, and so main conclusions could be drawn.

The place of the expert analyst (the human) is supposed and expected to be essential and major in that procedure. Instead of that the reality is quite different. It is important here to distinguish between Data Mining process (DMP) and human-machine interaction (HMI). [22] and [23] has shown that DMP is rather based on task-oriented systems whereas HMIs are limited to define them. The role of human is then divided into specific functions for a definitive application. These specific functions represent the basis of the knowledge system. The accomplishment of a task achieves the specific objective associated. This function can be subdivided according to its specificity. At this point the intervention of the expert may be determinant or even necessary. The main purpose here is to realise a tool capable to determine the duration of the episode of unemployment according to the profile of the candidate and aiming to forecast and estimate that duration. The major goal is to reduce the flow of jobless people and help the public intermediary to take immediate decision able to reduce their duration thanks to the use of simulator of unemployment.

To establish rules on which the simulator should act to correctly determine the duration of unemployment and individual recommendations, we have to determine in advance a model that will express the best the individual behavior of the tunisian unemployed according to their characteristics and attributes.

My initial base over the period from January 2010 to September 2015 includes, after filtering outliers, 206.409 male registrants (49.7%), 208.156 female registrants (50.3%). There are also 323.3667 unmarried (78%), 78.803 married (19%), 12.442 widowed (3%) and 37.328 divorced (9%). Our territorial database has 24 governorates (which will eventually go to 7 basins) and 256 delegations. There is also the presence of 75 types of initial diplomas (they will follow a grouping in 5 modalities) and 6 levels of schooling. It also includes registration dates and position dates (less than 60 days), registered unemployed persons (active and passive individuals).we do classified individuals by date of registration, date of position, age, sex, governorate (area), delegation (district), marital status, level of education, diploma, and specialties (other details were eliminated for the purposes of analysis, and an intelligent choice of most interesting and useful and discriminating[6] variables).

Then in order to get the best insights from qualitative variables that form the basis of my work on duration of unemployment, a Multiple Factor Component Modelling (MCA) is the best model that comes to mind [24].

The evolution of econometrics has made possible to go from an aggregated macroeconomic analysis of data to a microeconomic one of the individual attributes thanks to the development of computerized data processing and the various savings in time and money it allows[25][26]. Thus, in addition to the traditional quantitative statistical data that are usually processed, there is now a treatment of qualitative, more complex and heterogeneous characteristics (gender, socio-professional category, geographical affiliation, type of education, Contrary to being unemployed, etc.).Initially, traditional statistical methods only allowed modelling and analysis of quantitative characteristics. More specific methods have been developed and used since, allowing to take into account the absence of continuity of the variables treated, or the absence of a natural order between the modalities, in its qualitative dimension [27][28]. It is these specific methods that will be the subject of my work on the interest and the statistical significance of the qualitative variables, often neglected or omitted.

Historically, the study of models describing the modalities of one or more qualitative variables began in the 1940s. The most relevant research was Berkson's (1944-1951) research, including simple dichotomous models known Name of logit and probit models. The first empirical validations concerned mainly various sciences ranging from sociology to psychology through physics. It was not until the 1970s that the first attempts of these models in economics and political science were made, thanks in particular to the articles by MacFadden (1974) and James J. Heckman (1976).The modelling proposed by these authors has provided a framework for the application of econometric techniques of qualitative variables for the resolution of economic problems. This has made it possible to improve the interpretation of simple models of use and information and Synthetic material (logit).It was even later developed a mid-qualitative and semi-quantitative intermediate model (Tobit model), of interest and a certain contribution, for complex and diversified problems.

Thus a qualitative character that can take K different modalities. When K takes the modality 2, the latter is called dichotomous. Example: to be unemployed or not to be unemployed. In the general case, where K can take a number of modalities greater than 2, the variable is then called polyatomic. It is at this level that the difficulty of modelling econometrically a qualitative variable of this kind appears. Hence the advantage of switching to discrete-choice modelling allows us to answer and find detailed analytical solutions to our complex problem of duration. Then we will proceed to the application of the multi-mapping analysis MCA with the aim of having a first idea on the grouping of variables to be associated with the duration. We will then continue with a modelling in multinomial logit (ML) to try to express the duration of unemployment according to the different modalities, which explain it in probabilistic terms, which should allow us to determine the weight of the modalities (explanatory variables) in the duration of unemployment (explained variable).

Therefore we will go further trying to aggregate (ML)in to ordinal logit (OL) trying to express the unemployment duration according to the different modalities of the variables which explain it the best in probabilistic terms, and thus be able to compare the different results of discrete-choice modelling of qualitative variables.

 

Our sample is limited to the observation period from 1^st January 2006 to 30September 2015. As a result, some work stoppage may not be observed in full. A second censorship concerns input and output flows during the study period and should be taken into account in the analysis and modelling. Considering only the observed data in their entirety, is not the best solution since censors and truncations affect the likelihood, and the estimated parameters. The estimators would then be biased.

Figure 1. censorship and truncation

 

Intuitively we can say that if we consider censorship as a time of survival, the law of maintenance will be underestimated. As well if we omit truncation exit rates will be overestimated.

 

Figure 2. The most correlated variables to the unemployment duration

 

We do proceed to the MCA model in order to have an initial idea about the correlated qualitative variables (correlation ratio of the variables that make up the dimensions of the MCA). The ultimate goal is to find the variables that are the most correlated (close) to the duration (graphically) and therefore which could explain it the best. This is, however, not very statistically significant (we speak of individual qualitative variables) and will require a shift logit model to confirm or invalidate the results found.

We will thus review the representations of the active variables (formative and discriminating variables of the modelling) to that of the qualitative illustrative variable (the duration of unemployment for our case).Thus, what is taken from this figure:

• Gender and marital status are not graphically close to the duration, and thus a priori do not fall into the discrimination of the latter.

• Respectively, the governorate of residence (and / or membership of the candidates) and age (in the first place) appear to be the variables most correlated with duration.

• Level and diploma are also quite explanatory, and therefore related to duration, but of second importance in terms of proximity.

This figures although quite interesting remains incomplete and should be verified by an appropriate modelling for this type of variables (qualitative). The model that seems the most recommended is that of modelling the qualitative variables in discrete choices. This is reflected in some research that has attempted to model at best the transitions and probabilities of transitions in the job market. The synthesis of all these works, and in particular the work of Daniel L. MacFadden (1974) and James J. Heckman (1976)[29][30] allow us to retain two types of modelling which should allow us to represent and discriminate the duration of unemployment: the logit or the probit. A strong similarity between the two interaction models has been known for some time (Anas 1975 and Williams 1977) but the magnitude of this similarity has been underestimated. Our work here will try to prove that the two approaches are identical from the point of view of the results of the modelling. In the multinomial logit model we could obtain different modelling for different levels of modalities. The only concern is that one cannot aggregate towards a single modelling, or the statistical literature, remains silent on this problematic of aggregation, or even that excludes aggregation this type of modelling. In this context, logit modelling (less constraining and more general than probit modelling) is proposed for qualitative variables that should be coded because of the highly heterogeneous nature of the variables and modalities that need to be coded appropriately. From this we can compare a modelling with several modalities of the variables (multinomial) and a more aggregated modelling (ordinal) to draw the analyzes and conclusions.

 

E. Switching to the multinomial logit (ML)

 

The modelling in multinomial logit under SPSS 21 and Sub R converges towards a significance of all variables explaining the duration of unemployment, with the exception of marital status. Thus, age seems to be the most important variable that explains duration (as evidenced by the highest chi-square) followed by gender, the governorate of residence (area), diploma, year of diploma and specialties. This first important result tries to discriminate duration according to the other variables characteristic of unemployment, is a an additional confirmation of what has been advanced by number of authors having treated the subject. Indeed, the table in appendices provides various and varied modalities or the significance of the modalities of variables varies, with the exception of the variable of the marital status, which appears not significant, whatever the modality. The first model allows us to have an initial idea about the discriminating qualitative variables of the unemployment duration. However, the various modalities used do not allow a generalization and aggregation. This empirical constraint imposes a transition towards a more adapted and more aggregated modelling: the ordered logit.

Modelling an ordered logit is another model chosen by many authors of discrete choice models with respect to the modelling of qualitative variables[7]. The application of the ordinal logit to our database will allow us to obtain, contrary to the previously calculated multinomial logit, an aggregated modelling of the significant variables that affect the duration of unemployment. We also choose to join the 24 governorates to 7 regional areas in order to avoid the Multicollinearity problem (areas of east and west of north, center and south). We begin first by applying the tested logit (10% of the global base) and then to its application on the basis of its entirety in order to be able to verify and analyze the results which result from it.

Table2.  Results of sample test in ordinal logit model

\(Yduration=-0.4649897Xsex+0.018517XAge+0.3296339XSecondary+0.4270708XSecondaryprofessional+0.9734383XHigher+0.2467303XEastCenter-0.1452191XwestCenter-0.382215XwestNorth-0.7520418XEastSouth\)outh

The aggregation through the sample tested ordinal logit is :

 

 

 

As shown by the results of the modelling under STATA 12 for the test sample, the duration is significantly and positively correlated with all the variables of the base, except for the feminine gender, the belonging to the Center west, North west and South west regional areas with which the duration is certainly significant but negatively correlated. The only exception is marital status (married) which is not significant and discriminated to duration. The duration of unemployment is inversely proportional to female candidates. This is a common feature for many underdeveloped countries such as ours, as reported by few publications or articles on the subject [31]. Second observation concerns the belonging of the unemployed candidates to Central West, North West and South East. This is quite predictable for these regions of western and southern Tunisia which have been historically marginalized for decades, testifying about the strong attachment of the unemployed to these regions (West and South) as well as showed by the index of regional development of the country far from the rich and prosperous north and central regions. The explanation of these results could be related to the low population of these regions, the development of the black market with Algeria on the one hand and Libya on the other, the low rate of urbanization (large governorates such as Kebili for example with a small number of inhabitants), the lack of employment offices, the discouragement of young people, the low rate of direct investment, the lack of State programs and so on. In addition to that the modalities for marital status are positively and significantly correlated with duration. Especially the single, widowed and divorced status which rises with duration, contrary to the married modality. Thus, it seems that the fact to be married does not help to find a job more easily.

For our database, we were surprised by the fact that despite that the majority of registered unemployed are illiterates, this modality is not correlated or does not discriminate the duration of unemployment. Regarding the variable level of schooling, all the modalities are significantly and positively correlated, and especially the higher level. This result converges with the official statistics of the country (NSI and NEASE) and most international organizations (UN, UNDP, OECD) .This feature, common once again to less developed countries, is not very discriminatory in most developed countries. Again, a significant gap in the analytical spirit emerges from research on developed countries, which categorically diverge from those of the less developed countries in the results, analyzes and reasons [32][33][34].

After the tested logit we do generalised the analysis to our entire database and then we obtain practically the same results:

 

 

Table3.  Results of ordinal logit model

We find practically the same signs for the most discriminating variables retained by the modeling test in ordered logit:

 

 

 

Thus, to the significant variables retained by the test sample, are added other significant and discriminating variables of the duration, in particular the year of graduation and especially the regional affiliation of the candidates (which is significant and negative). If we add the diploma variable, which is not significant (problem of multi-collinearity with the specialty and the level of schooling) year of graduation (which is significant and negative), we will obtain this stable model for the ordinal logit which converges with the result of the test sample.

 

 

 A decision support system (DSS) in which we have included prior probability distributions [35] and the SHARK (Search Hierarchic Association Rules for Knowledge) algorithm [36] is applied. For a better interaction with the user who has the task to direct the search, an easy-to-use interface has also been developed. This tool is a necessary support to help the experts in their analysis. Indeed, this software could be programmed with the best filtered data and better representative algorithms but remain sometimes unsuitable, despite of this, because of the difficulty and complexity of the matter in hand. Therefore it is necessary to work bearing in mind that the final application (the soft) should be in accordance with the expectations and aspirations of users. We used a specific graphical interface to be able to program and increment our algorithm rules relating to duration. The principle is based on the anthropocentric approach.

 

 

Figure3. Data Mining Process and work flow[8]

 

The guiding thread allows reducing the expertise times and the number of association rules generated by this specific procedural method by including the user-analyst at the heart of the Data Mining process. Hierarchical search is used to perform searches level in order to quickly propose generalized rules to a user who will judge their relevance. His choices will guide the process in the following levels to specify the rules. In that way, the number of rules generated is smaller and more targeted because the user guides the search from the beginning to the end of the process. A set of algorithms has been developed to try to meet these specifications.

For that goal we realised is a multipurpose tool: a simulator. It first helps in calculating the average of unemployment duration for each subscripted candidate, his probability of unemployment and mainly a decision- making tool able to reduce the time wasted to get a job, and consequently increasing the probability of leaving unemployment. The static database on which we focused our efforts is covering the period 2005 till 2015 for individuals already registered. We faced the classic problems of statistical censorship and truncation, for which it was possible to overcome through incremental updates to the system. This is one of the advantages of intelligent self-learning systems.  Self-learning allows the simulator to take into account the instantaneous changes (numbers and profiles) affecting the base (hence self-learning) and thus help to extract realistic rates, durations and decisions that may affect unemployed persons. All this required a transition to computer programming through the Access database to correctly define the scripts and algorithms that may correctly establish and express relations and operations dealing with set of data (or recommendations).These methods allow us to simulate the processes of human reasoning (inference, analogy and deduction) based on the available basic knowledge[37][38]. Another very important aspect of developing intelligent systems is their ability to acquire new knowledge (sometimes from several different sources) and evaluating them. Then, we  tried to answer to the major problem of research :”how to manage the individual determinants of the duration of unemployment in Tunisia” based on hybrid systems combining a database and a set of interconnected algorithms, established under an intelligent and decision-making simulator combined with the human expertise granted by the agents of NEASE. The information collected through the NEASE[9] network is then processed by a learning process. This treatment is then ensured both by updates, and also correction of the gaps or inconsistencies thanks to the use of a constructive network. A method of extraction described as incremental rules were integrated into the system, together with knowledge validation algorithms, marrying connectionist and symbolic modules [40]. For all that, the simulator was designed to generate an automatic learning system for Constructive (incremental) acquisition of knowledge [41][42]. All this laid the foundation of the context in which our application on the tunisian labour market is made, aiming to take in account ever-increasing number of applicants, the absence of a personalized follow-up, precise and detailed ratios or indicators of this market, the individual characteristics of the applicants, the diversity of inadequate offers. The public intermediary is now given all the means necessary to carry out its duties. Facing the obligation of playing perfectly its role of consultant and find ways to determine matching and to control it. The offices of this Agency have on their possession at this time an instant dashboard to monitor the situation of the unemployed and are able to advise them, and supervise the effectiveness of corrective measures of duration and hence of unemployment .It is in this sense and for those goals, that the idea of a simulator of the duration of unemployment was made.

Concretely, we will be able to calculate first of all, the instantaneous duration of the unemployed enrollers via a simulator of the duration of unemployment, taking into account the individual specifications and the rules of association established through our initial database. We can then increment any new registrant to automatically update the database and have accurate and unbiased calculations. A last step, and not the least, we will present the recommendations of this simulator to reduce the duration for each of the registrants according to their personal variables and attributes. This will be preceded by the integration of a data set or physical schema of the simulator.

Figure 4. Physical relational schema of the simulator

 

Finally we will have a decision support computer system (DSCS) for which we proposed the following graphical interface.

 

Figure 5. Graphic Interface of the simulator

 

3  Primary analysis of the unemployment duration and simulator implementation

The primary analysis of the unemployment duration at the level of the various governorates reveals an imbalanced distribution. Far from the classic observation which attests that coastal areas (rich and prosperous in theory) are not affected by unemployment and therefore by the duration associated with it, we found on the contrary, confused and sometimes contradictory relationship between theory and official statistics.

Figure 6. Distribution of unemployment duration in Tunisia

It is already noted that the zones are classified into four groups:

 

• The most affected by a high duration of unemployment are the regions of Tunis, Monastir, Mehdia, Sfax and Tozeur, where the average duration varies between 130 and 142 days.

• Follow Bizerte, Sousse, Mannouba, Ariana, Sidi Bouzid and Gafsa with a waiting time average between 123 to 129 days.

• Beja, Zaghouane, BenArous, Nabeul, Silena, Kasserine and Kébili governorates lasting between 113 and 123 days.

• Follow Jendouba, Kef, Kairouan, Tataouine, Medenine and Gabes for an average time between 96 and 112 days.

 

Thus the existence of this macroeconomic imbalance in the distribution of duration already determines the framework within which the microeconomic analysis that we are about to realize through the simulator.

 

 

The association of the various criteria (and / or variables) of our initial database allows us to have a double result: first, the instantaneous individual duration of each candidate according to the governorate of belonging, gender, marital status, age, diploma, and specialty. Then, the simulator determines the number of cases concerned, by such criteria in terms of population registered in the employment offices. Thus, by simulating the average duration of the population of registered unemployed male for example, without assigning precise selection criteria, we obtain an average duration of 108 days for 201,835 individuals.

 

Figure 6. Average of duration and number of cases

 

After that if we took randomly the governorate of Ariana as affiliation criteria, for example, in addition to male candidate and unmarried marital status, the average unemployment duration of 132.24 days is obtained, for 6407 registered unemployed.

 

Figure 7. Gender and residence affiliation

 

This first overview of the simulated duration of unemployment is followed by a series of recommendations suggested by the simulator to reduce this duration and to be able to improve the probability of unemployed people leaving unemployment. The aim is to reduce this duration of unemployment and thus to be able to retroact (active employment policies).

The simulator presents, first, the recommendations on geographical mobility with particular interest to the areas to which the unemployed belong. An increasing chronological classification of the duration, the districts of the same area but also the closest geographically area is then illustrated.

 

Figure 8. Duration for the same candidate profile at the national level

 

In addition to that, a second set of recommendations is proposed by the simulator according to the study specialties for the jobless whom waiting times are the longest. Their probability of leaving unemployment is then the highest. The simulator gave a particular interest to the area to which the unemployed belong, to propose to them the district that may match the most to their profiles, their specialties and their vocational training ( smallest duration).

 

Figure 9. Chronological recommendation by speciality

 

From there, the two sets of recommendations can be combined to finally recommend the vocating training and districts, in line with the candidate's attributes or determinants.

 

Figure 10. Spatial recommendation of the simulator

 

We are then facing a new tool to assist decision-making, able to help labour market intermediaries and in particular the NEASE agency, to fulfil their role as consultant and advisor. As much as the number of criteria associated to the candidate increases, more efficient and accurate the calculations and suggested measures becomes.

The role that has been assigned to NEASE agencies till nowadays, limited to a simple data collector, neutral most of the time, and incapable of making decisions or advice jobseekers, has now all the chances and opportunities to change ensuring  the real function for which it was created .This is the advantage of this innovative tool for the intermediary. The simulator of unemployment duration should make possible for the NEASE offices to have an instant dashboard to monitor the situation of the unemployed and to be able to advise unemployed and to supervise the correctives measures for efficiency and by the same way to contribute to reduce unemployment. This application will provide job seekers and companies the means to be served by any employment office in Tunisia. The centralized database is the result of centralising all data bases of actors of job market, which are instantly updated. This dashboard is representative of all evolutions and the related calculations are fairly accurate and credible. Individual information on demand and offer, as well as another indicator are provided by representations of NEASE.

In fact, the simulator provides the managers of the employment database the ability to increment it at any time from any office thanks to password and a login belonging to the accredited agents.

 

Figure 11. Management of Database

 

 

 

This new computerized decision-making tool provides consequently a source of both personal and global information for whom in charge of managing the unemployment problem, the NEASE agencies in particular. Indeed, the duration of unemployment is indicative of the length of episodes of unemployment for a candidate with a profile and a set of particular characteristics and attributes. Thus, taking into account the variables of the individuals registered at the employment offices and applying the associative algorithms established by the simulator, it is possible to determine precisely for a particular profile the exact duration, number of cases involved etc.

In a second step, the simulator establishes a national scale at the level of the 7 areas, 24 districts and 256 delegations, the duration of unemployment with distinctive criteria.

In a final step, the simulator will fully play its decision-making role by proposing a shorter duration for a geographically adjacent (spatially contiguous) district belonging to the overall area of the candidates. At the same time, the simulator also offers for diplomas and study specialties in chronological order for duration, both to allow having an idea about the training courses most requested and recommended spatial weakest duration associated.

This should make it possible to in better advices for candidates, in search of employment associated to their migratory desire correlated to the vocating training they have for a better probability of leaving unemployment. This should enable policy-makers, on the whole,

 to determine the training that generates jobs, and which should be reduced or eliminated.

At the same time, this should enable us to achieve two objectives: one at the individual level (exit from unemployment) and the other at a more macroeconomic and even strategic level (in better matching between job offer and demand).Also, this practical and innovative tool also has the advantage of being incremental, meaning that any new registration will be automatically counted and the calculation of the duration will be completely up to date.

The management of individual determinants of the duration of unemployment will now be implemented by NEASE and its agents. They can thus have at any time and in any office, the situation of all the candidates according to their various characteristics or attributes. In light of the information gathered, they can inform, guide, advise and analyze thereafter at real-time and insure an efficient management.

 

[1] Autor, D. H. (2001). Wiring the labor market. The Journal of Economic Perspectives, 15(1), 25-40.

[2] Bessy, C., & Chauvin, P. M. (2013). The power of market intermediaries: From information to valuation processes. Valuation studies, 1(1), 83-117.

[3] Autor, D. (2008). The economics of labor market intermediation: An analytic framework.

[4] Mortensen, D. T., & Pissarides, C. A. (1999). New developments in models of search in the labor market. Handbook of labor economics, 3, 2567-2627.

[5] George J.Stigler :Information in the Labor Market, Journal of Political Economy.Vol. 70, No. 5, Part 2: Investment in Human Beings (1962),94-105, Published by The University of Chicago Press.

[6] Miller, M. H., & Rock, K. (1985). Dividend policy under asymmetric information. The Journal of finance, 40(4), 1031-1051.

[7] Spence, M. (1973). Job market signaling. The quarterly journal of Economics, 87(3), 355-374.

[8] Guillemette de Larquier : Principes des marchés régis par appariement, Revue économique.Vol. 48, No. 6,1409-1438,(1997),Published by: Sciences Po University Press.

[9] Jovanovic, B. (1979). Job matching and the theory of turnover. Journal of political economy, 87(5, Part 1), 972-990

[10] Bleakley, H., & Fuhrer, J. C. (1997). Shifts in the Beveridge curve, job matching, and labor market dynamics. New England Economic Review, 3.

[11] Noe, R. A., Hollenbeck, J. R., Gerhart, B., & Wright, P. M. (2006). Human resource management: Gaining a competitive advantage.

[12] Bizer, C., Heese, R., Mochol, M., Oldakowski, R., Tolksdorf, R., & Eckstein, R. (2005). The impact of semantic web technologies on job recruitment processes. In Wirtschaftsinformatik 2005 (pp. 1367-1381). Physica-Verlag HD.

[13] Rumberger, R. W., & Levin, H. M. (1985). Forecasting the impact of new technologies on the future job market. Technological Forecasting and Social Change, 27(4), 399-417.

[14] HOUTSMA, Maurice et SWAMI, Arun. Set-oriented data mining in relational databases. Data & Knowledge Engineering, vol. 17, no 3, (1995), 245-262.

[15] KODRATOFF, Yves. Applications de l'apprentissage automatique et de la fouille de données. In : EGC.(2001),57-68.

[16] Piateski, G., & Frawley, W. (1991). Knowledge discovery in databases. MIT press.

[17] Brin, S., Motwani, R., Ullman, J. D., & Tsur, S. (1997, June). Dynamic itemset counting and implication rules for market basket data. In ACM SIGMOD Record (Vol. 26, No. 2, pp. 255-264). ACM.

[18] Li, L., & Zhang, M. (2011, July). The strategy of mining association rule based on cloud computing. In Business Computing and Global Informatization (BCGIN), 2011 International Conference on (pp. 475-478). IEEE.

[19] Leung, M. D. (2014). Dilettante or renaissance person? How the order of job experiences affects hiring in an external labor market. American Sociological Review, 79(1), 136-158.

[20] Poterba, J. M., & Summers, L. H. (1995). Unemployment benefits and labor market transitions: A multinomial logit model with errors in classification. The Review of Economics and Statistics, 207-216.

[21] WILLIAMSON, Oliver E. Comparative economic organization: The analysis of discrete structural alternatives. Administrative science quarterly, (1991), 269-296.

[22] Elton, M. D., & Book, W. J. (2010). Operator efficiency improvements from novel human-machine interfaces. Georgia Institute of Technology.

[23] Takagi, H. (2001). Interactive evolutionary computation: Fusion of the capabilities of EC optimization and human evaluation. Proceedings of the IEEE, 89(9), 1275-1296.

[24] Abdi, H., & Valentin, D. (2007). Multiple correspondence analysis. Encyclopedia of measurement and statistics, 651-657.

[25] Pindyck, R. S., & Rubinfeld, D. L. (1991). Econometric models. Economic Forecasts, 3.

[26] McFadden, D. L. (1984). Econometric analysis of qualitative response models. Handbook of econometrics, 2, 1395-1457.

[27] Gujarati, D. N. (2009). Basic econometrics. Tata McGraw-Hill Education.

[28] Phillips, P. C., & Sul, D. (2007). Transition modeling and econometric convergence tests. Econometrica, 75(6), 1771-1855.

[29] Daniel MacFadden « Conditional Logit Analysis of Qualitative Choice Behavior » 1974- Major curse At University of California - Barkeley

[30] Flinn C, Heckman J. (1983). "Models of the Analysis of Labor Force Dynamics", in Advances in Econometrics, vol.I, R. Basmann et G. Rhodes éditeurs, JAI Press, Greenwich, pp. 35-95.

[31] Mark C.Foley de l’université de Yale « Determinants of unemployment duration in Russia » Aout 1997- Economic Growth Center

[32] Erna Kahraman Phd june 2011 ˝Youth Employment and Unemployment in Developing Country : Macro challenges with Micro Perspectives ˝.

[33] McCormick B. (1990). "A Theory of Signalling During Job Search, Employment Efficiency, and Stigmatised Jobs" , The Review of Economic Studies, vol. 57, pp. 299-313.

[34] Maki D.R., Spindler Z.A. (1975). "The Effect of Unemployment Compensation on the Rate of Unemployment in Great-Britain", Oxford Economic Papers, vol. 27, n°3,pp. 440-454, novembre.

[35] Paul Gregg,Jonathan Wadsworth : How effective are state employment agencies ? Jobcentre use and job matching in Britain.Oxford Bulletin of Economics and Statistics,Volume 58, Issue 3,  443–467, (1996).

[36] AGRAWAL, Rakesh, IMIELINSKI, Tomasz, et SWAMI, Arun :Database mining: A performance perspective. IEEE transactions on knowledge and data engineering, (1993), vol. 5, no 6, 914-925.

[37] GREENBERG,Jerald et BARON,RobertA:Behavior in organizations: Understanding and managing the human side of work ,2003,Pearson College Division.

[38]  DIAPER. Dan : Understanding task analysis for human-computer interaction. The handbook of task analysis for human-computer interaction, (2004), 5-47.

[39] HAN, Jiawei et KAMBER, Micheline : Data mining: concepts and technologies, Models Methods & Algorithms, (2001), vol. 5, no 4,1-18.

[40] SRIKANT, Ramakrishnan et AGRAWAL, Rakesh. Mining sequential patterns: Generalizations and performance improvements. In : International Conference on Extending Database Technology. Springer Berlin Heidelberg,(1996),1-17.

[41] HOUTSMA, Maurice et SWAMI, Arun :Set-oriented data mining in relational databases. Data & Knowledge Engineering, (1995), vol. 17, no 3, 245-262.

[42] DIAPER, Dan et SANGER, Colston : Tasks for and tasks in human–computer interaction. Interacting with Computers, (2006), vol. 18, no 1, 117-138.

 

 

· Ethics approval and consent to participate

· Consent for publication

· List of abbreviations

NEASE = National Employment Agency and Self employment.

IT = Information technologies.

CSDM = Computer system for decision-making.

KDD =Knowledge Discovery in Databases.

DMP = Data Mining process.

HMI = human-machine interaction.

NIS/NSI = National Institute of statistics of Tunisia.

MCA = Multiple Factor Component Modelling.

ML = Multinomial Logit.

OL = Ordred Logit.

DSS = Decision Support System.

SHARK = Search Hierarchic Association Rules for Knowledge.

· Availability of data and material

The data of the unemployment duration of individuals in Tunisia was sent to me by Mr Mohamed Fadhel Berhouma Director Of  Information Systems and managment of Method on National Employment Agency and Self employment in Tunisia.

Adress :19 Rue Asdrubal 1002 Tunis - Tunisie

Tel.: +216.71.781.200  / Fax : +216.71.783.236

Email: fadhel.benrhouma@gmail.com

The data can be download through :https://drive.google.com/file/d/0B2IWlb_9qlhYLXdCNVNhNnJHU00/view

 

· Competing interests

‘No competing interests by this reserch’

· Funding

‘No funding used for this reserch’

· Authors' contributions

To answer the main problem of this research paper (management of the individual determinants of the duration of unemployment) i have tryed to explain:

ü Intermediation and employment policies

ü  The realization of a software of calculation of duration of unemployment (Simulator)

ü  The redefinition of the role of the intermediary in Tunisia

The main contributions could be summarized in 5 points:

• A new computerized decision support tool providing information for both registered unemployed persons and organizations responsible for managing the unemployment problem, in particular the NEASE .

• Using this new soft, it would be possible to determine for a particular profile of candidates the exact duration and number of cases involved. And on a macroeconomic scale, the duration of unemployment for the same profile of candidates nationwide.

• It also offers diplomas and study specialties, the duration of which is arranged in chronological order so that one can get an idea of the most requested and lasting diploma courses. This should make it possible to better advise candidates in search of employment in order to associate their migratory desire with the training they have for a better probability of leaving unemployment.

Ø This makes it possible for policy-makers to determine the training that generates employment and which should be reduced or eliminated.

• This tool is incremental, so any new enrollment will be counted automatically and the calculation of the duration will be updated.

• Management of the individual determinants of the duration of unemployment will now be implemented by NEASE and its regional offices. They can thus have, at any time and in any office, the detailed situation of all the candidates, according to their various characteristics or attributes. They can therefore inform, guide, advise and analyze. They will then play the main role for which they were created with a more effective and instantaneous management of the registered unemployed.

· Acknowledgements

- Mr. Ricco Rakatomalala, lecturer at the University of Lyon 2.

- Mr. François Husson, Professor at Agrocampus-Ouest and Director of the IRMAR laboratory.

- Mr. William Greene of the University of New York Stern School of Business.

- Mr. Adel Ben Rhouma, Director of IT, at the National Agency for Employment and Independent Labor.

· Authors' information (optional)

Anis Ben Ahmed Lachiheb

Adress : Résidence les 3 Palmiers –

Avenue Abdlahmid Essaka-

Bouhsina- Sousse (4000) (TUNISIA)

IHEC Sousse (TUNISIA)

Adress : Route Hzamia Sahloul 3 –

 BP n° 40 - 4054 Sousse - 4054 sousse

Tel : +216 73 368 351 - 73 368 358

Fax : +216 73 368 350

 

Corresponding author : Anis Ben Ahmed Lachiheb

Classification JEL : R 23, R 19, R 38