|
|
 |
| |
|
|
|
A Comparative Analysis of Decision Trees, Bagging, and Random Forests for Predictive Modeling in Monetary Poverty: Evidence from Morocco |
|
|
|
PP: 233-240 |
|
|
|
doi:10.18576/amis/180203
|
|
|
|
Author(s) |
|
|
|
El aachab Yassine,
Kaicer Mohammed,
|
|
|
|
Abstract |
|
|
| Predicting monetary poverty is important and has broad effects on social and economic growth. In this field, precise and useful predictive modeling is essential because it helps humanitarian groups and policymakers allocate resources more effectively and focus interventions more effectively. We present a thorough comparison and examination of three different machine-learning approaches: Random Forests, Bagging, and Decision Trees. Our main objective is to assess their effectiveness and suitability in the particular context of forecasting monetary poverty in the Moroccan region. We begin with Decision Trees, which are renowned for their openness and interpretability. Although they provide a clear understanding of the decision-making process, their prediction accuracy may be limited. In order to improve prediction accuracy, we investigate the potential of Bagging, a combination method that aggregates several Decision Trees. We also explore the more sophisticated ensemble method of Random Forests, where higher robustness and performance are anticipated due to the randomness introduced in feature selection during tree construction. We use real-world datasets that are closely associated with financial poverty in our investigation. We carefully assess each methodology’s computing efficiency, model resilience, and forecast accuracy. Additionally, we explore the effects of hyperparameter tuning, feature engineering, and the specific properties of the dataset on our results. The models’ outputs are assessed using a number of measures, including accuracy, precision, Cohen’s Kappa statistic, F1-score, and recall. The R values show that all three algorithms had very good accuracy ratings. As a result, the accuracy of the Bagging approach is higher (99.94%) than that of the Random Forest and decision tree methods (99.61%) and (98.45%). Through this research, we endeavor to unearth insights into the strengths and limitations of these machine- learning techniques in the context of monetary poverty prediction. The knowledge garnered from this study is poised to offer invaluable guidance to decision-makers and researchers alike, as they address the intricate challenge of predicting and mitigating monetary poverty in the Morocco region.
|
|
|
|
|
|
|
|
