The Comparison of Machine Learning Model to Predict Bankruptcy: Indonesian Stock Exchange Data
Ednawati Rainarli, M.Si.
Universitas Komputer Indonesia
Abstract
This research focused on determining the machine learning model that used to predict bankruptcy. It got the data from the financial statements of public companies; reported by the Indonesian Stock Exchange from 2009 until 2015. The predictions of the bankruptcy used the training model of machine learning method. The analyzer features were the accounting ratios; used in a statistical analyzing of the financial statement. Handling missing values, choosing correlation features that related to classes, and dealing with imbalanced dataset were some problems have been solved at the beginning of the preprocessing phase. The training process used the preprocessing result to fit the data with the prediction model. Accuracy is used to measure the performance of the models in predicting bankruptcy. The results are Sequential Minimal Optimization (SMO) with Linear kernel functions works the best to predict 1-year before bankruptcy with the accuracy of 91.57% and SMO with Radial Basis Function (RBF) works well to predict 2-years before bankruptcy; the accuracy is 93.8%. This study has shown the influence of feature selection and normalization processes in making precise predictions and has shown SMO as a potential method to make predictions.
Keywords: prediction, bankruptcy, accounting ratio, machine learning, missing value, feature selection, imbalanced dataset
Topic: Electrical and Computer Engineering