Research Article
Loan Status Prediction System with Ensembled Machine Learning Models: Elevating Information Reliability and Accuracy
@INPROCEEDINGS{10.1007/978-3-031-66044-3_19, author={K. Badri Narayanan and Yagnesh Challagundla and Dev Rishik Maruturi and Nihar Ranjan Pradhan}, title={Loan Status Prediction System with Ensembled Machine Learning Models: Elevating Information Reliability and Accuracy}, proceedings={Pervasive Knowledge and Collective Intelligence on Web and Social Media. Second EAI International Conference, PerSOM 2023, Hyderabad, India, November 24--25, 2023, Proceedings}, proceedings_a={PERSOM}, year={2024}, month={8}, keywords={Machine learning RanNeu Loan Prediction Embedded modeling Prediction system Data Visualization}, doi={10.1007/978-3-031-66044-3_19} }- K. Badri Narayanan
Yagnesh Challagundla
Dev Rishik Maruturi
Nihar Ranjan Pradhan
Year: 2024
Loan Status Prediction System with Ensembled Machine Learning Models: Elevating Information Reliability and Accuracy
PERSOM
Springer
DOI: 10.1007/978-3-031-66044-3_19
Abstract
We aimed to develop an integrated tool to more reliably and consistently anticipate lending conditions. The dataset for our analysis, which included a number of feature categories, was offered by Elsevier. The dataset had been uploaded to the application and then preprocessed. At this stage, we encountered missing data and found numerous extra columns, which we promptly eliminated. By eliminating pointless or duplicate features, we hoped to increase the model’s ability to draw out important patterns from the data. Furthermore, we handled missing data by transforming discrete variables and imputing with average/most common values. The dataset was subsequently split into training and testing sets using a 70:30 ratio, and 5-fold cross-validation was used to evaluate the data. We examined a number of techniques for machine learning before choosing on Neural Networks, Gradient Boosting, Random Forest, and an innovative algorithm we created termed “RanNeu” (an Embedded ML model fusing Random Forest with Neural Networks). We carefully selected hyperparameters for each machine learning model in order to maximize performance. Using Lasso (L1) regularization, a constant learning rate, and an initial learning rate (eta) of 0.0100, we improved Gradient Boosting. As a result, we increased the number of neurons in hidden layers for neural networks to a maximum of 300, employed ReLu activation, and applied the Adam solver approach while using regularization with an alpha value of 0.0001 for neural networks.
