Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Predict Saving Behavior - Artificial Neural Network & Machine Learning

Möllestam, William LU (2022) NEKP01 20221
Department of Economics
Abstract (Swedish)
This study aims to predict saving behavior using Artificial Neural Network (ANN), XGBoost, and Support Vector Machine (SVM) algorithms. First, 25 variables were chosen from the original 217 questions asked by the National Financial Capability Well-Being Survey (2018) NFCS, using exploratory data analysis. K-means clustering was applied to determine the optimal number of saving classes (k=5) to include in the final model. Thereafter, a five-fold cross validation (CV) technique was used to tune each model's hyperparameters. Using the optimal hyperparameter configuration and a training set of 70% of the data, prediction models were constructed. The performance of each model was then evaluated using the test set (30% of the data). The... (More)
This study aims to predict saving behavior using Artificial Neural Network (ANN), XGBoost, and Support Vector Machine (SVM) algorithms. First, 25 variables were chosen from the original 217 questions asked by the National Financial Capability Well-Being Survey (2018) NFCS, using exploratory data analysis. K-means clustering was applied to determine the optimal number of saving classes (k=5) to include in the final model. Thereafter, a five-fold cross validation (CV) technique was used to tune each model's hyperparameters. Using the optimal hyperparameter configuration and a training set of 70% of the data, prediction models were constructed. The performance of each model was then evaluated using the test set (30% of the data). The precision, recall, and F_1 indexes were used to analyze the prediction performances of each saving class, whereas the accuracy and their macro-average values were applied to evaluate the overall performance of the prediction model. The relative importance of each variable was determined based on the sensitivity analysis of the variables. The financial planning horizon and how long individuals believed they would live had the biggest influence on prediction outcomes. In addition, classical economic methods and other ML algorithms were adopted as comparisons. The results showed that ANN, XGBoost, and SVM algorithm achieved a better comprehensive performance, and their prediction accuracies were 0.85, 0.84, and 0.80, respectively. For questions related to behavioral economics and saving behavior, the presented methodology can serve as a reliable reference. (Less)
Please use this url to cite or link to this publication:
author
Möllestam, William LU
supervisor
organization
course
NEKP01 20221
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Saving Behavior, Artificial Neural Network, Machine Learning, XGBoost, SVM
language
English
id
9100774
date added to LUP
2022-10-10 11:29:04
date last changed
2022-10-10 11:29:04
@misc{9100774,
  abstract     = {{This study aims to predict saving behavior using Artificial Neural Network (ANN), XGBoost, and Support Vector Machine (SVM) algorithms. First, 25 variables were chosen from the original 217 questions asked by the National Financial Capability Well-Being Survey (2018) NFCS, using exploratory data analysis. K-means clustering was applied to determine the optimal number of saving classes (k=5) to include in the final model. Thereafter, a five-fold cross validation (CV) technique was used to tune each model's hyperparameters. Using the optimal hyperparameter configuration and a training set of 70% of the data, prediction models were constructed. The performance of each model was then evaluated using the test set (30% of the data). The precision, recall, and F_1 indexes were used to analyze the prediction performances of each saving class, whereas the accuracy and their macro-average values were applied to evaluate the overall performance of the prediction model. The relative importance of each variable was determined based on the sensitivity analysis of the variables. The financial planning horizon and how long individuals believed they would live had the biggest influence on prediction outcomes. In addition, classical economic methods and other ML algorithms were adopted as comparisons. The results showed that ANN, XGBoost, and SVM algorithm achieved a better comprehensive performance, and their prediction accuracies were 0.85, 0.84, and 0.80, respectively. For questions related to behavioral economics and saving behavior, the presented methodology can serve as a reliable reference.}},
  author       = {{Möllestam, William}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{Predict Saving Behavior - Artificial Neural Network & Machine Learning}},
  year         = {{2022}},
}