Machine Learning Ensemble Modelling for Predicting Unemployment Duration

Abstract

Predictions of the unemployment duration of the economically active population play a crucial assisting role for policymakers and employment agencies in the well-organised allocation of resources (tied to solving problems of the unemployed, whether on the labour supply or demand side) and providing targeted support to jobseekers in their job search. This study aimed to develop an ensemble model that can serve as a reliable tool for predicting unemployment duration among jobseekers in Slovakia. The ensemble model was developed using real data from the database of jobseekers (those registered as unemployed and actively searching for a job through the Local Labour Office, Social Affairs, and Family) using the stacking method, incorporating predictions from three individual models: CART, CHAID, and discriminant analysis. The final meta-model was created using logistic regression and indicates an overall accuracy of the prediction of unemployment duration of almost 78%. This model demonstrated high accuracy and precision in identifying jobseekers at risk of long-term unemployment exceeding 12 months. The presented model, working with real data of a robust nature, represents an operational tool that can be used to check the functionality of the current labour market policy and to solve the problem of long-term unemployed individuals in Slovakia, as well as in the creation of future government measures aimed at solving the problem of unemployment. The measures from the state are financed from budget funds, and by applying the appropriate model, it is possible to arrive at the rationalization of the financing of these measures, or to specifically determine the means intended to solve the problem of long-term unemployment in Slovakia (this, together with the regional disproportion of unemployment, is considered one of the most prominent problems in the labour market in Slovakia). The model also has the potential to be adapted in other economies, taking into account country-specific conditions and variables, which is possible due to the data-mining approach used.

Artificial neural network and decision tree-based modelling of non-prosperity of companies

“…As the above-listed authors mentioned in their studies, the ANNs usually achieve a pretty high prediction accuracy, but on the other hand, the resulting models are considered black boxes, as they are so complex that they cannot be interpreted. Besides the ANNs, the DT-based models are very clear to understand but sometimes have quite lower classification ability (Gabrikova et al, 2023). However, as we intend to demonstrate in this study, this can be strengthened by employing ensemble modelling using various techniques for enhancing the accuracy or stability of the models.…”

Human Error Analysis and Fatality Prediction in Maritime Accidents

“…The principle of binary logistic regression is as follows [34]. Let Y be a binary dependent variable that takes the value 1 (fatality occurred) with probability p and the value 0 (fatality did not occur) with probability 1 − p. Let X 1 , .…”

“…If the causal factor was not present, its indicator variable had a value of 0. The value 0 was chosen as the reference category in each case for the better interpretation of the coefficients of the created model [34].…”

“…The quality of the model can then be quantified as the area under the ROC curve (Area Under the Curve-AUC). The closer this value is to 1, the better the model is as a classifier [34]. All calculations are performed using IBM SPSS Modeler software, version 18.…”