Analisis Data Panel Pada Kinerja Keuangan Menggunakan Metode Regresi Statistik, Machine Learning dan Deep Learning

Teguh, Andy Mohammad (2025) Analisis Data Panel Pada Kinerja Keuangan Menggunakan Metode Regresi Statistik, Machine Learning dan Deep Learning. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6025231047-Master_Thesis.pdf Restricted to Repository staff only until 1 April 2027. Download (4MB) | Request a copy

Abstract

Prediksi kinerja keuangan perusahaan merupakan bagian penting dalam pengambilan keputusan strategis, terutama bagi industri manufaktur. Namun, metode tradisional seperti regresi data panel memiliki keterbatasan dalam menangani kompleksitas data keuangan yang dinamis. Penelitian sebelumnya telah mengeksplorasi teknik machine learning dan deep learning, tetapi belum banyak yang secara langsung membandingkan efektivitas metode ini dengan pendekatan regresi statistik. Selain itu, peran feature selection dalam meningkatkan akurasi model prediksi masih kurang mendapat perhatian. Penelitian ini mengusulkan pendekatan integrasi feature selection dalam memprediksi Return on Assets (ROA) dan Return on Equity (ROE) perusahaan manufaktur di Indonesia. Penelitian ini melibatkan pra-pemrosesan data, pengujian stasioneritas dengan Augmented Dickey-Fuller (ADF), serta eliminasi variabel multikolinearitas tinggi menggunakan Variance Inflation Factor (VIF). Selain itu, pemilihan fitur dilakukan melalui Feature Importance Ranking (FIR) dan analisis p-value regresi statistik untuk meningkatkan efisiensi prediksi. Hasil penelitian menunjukkan bahwa pemilihan fitur pada skenario 4 dengan pendekatan integrasi VIF dan signifikansi p-value berhasil memperbaiki nilai Mean Squared Error (MSE) model Multilayer Perceptron (MLP) sebesar 96,11% dan 90,72% pada model Convolutional Neural Network (CNN) dibandingkan pada skenario 3 dalam melakukan prediksi ROA. Random Forest merupakan model dengan performa terbaik dalam memprediksi ROE dengan nilai MSE sebesar 4,31828e-06. Sementara itu, model Light Gradient-Boosting Machine (LightGBM) menunjukkan hasil terbaik dalam memprediksi ROA dengan nilai MSE sebesar 8,75833e-07. Model regresi statistik, khususnya Random Effects Model (REM), memiliki keterbatasan dalam menangani data dengan multikolinieritas tinggi. Dengan demikian, penelitian ini menegaskan bahwa integrasi feature selection memiliki peran penting dalam meningkatkan akurasi prediksi, serta bahwa model LightGBM merupakan model unggulan dibandingkan pendekatan statistik konvensional.
==================================================================================================================================
Predicting a company's financial performance is essential to strategic decision-making, especially for the manufacturing industry. However, traditional methods such as panel data regression have limitations in handling the complexity of dynamic financial data. Previous research has explored machine learning and deep learning techniques, but not many have directly compared the effectiveness of these methods with statistical regression approaches. In addition, feature selection's role in improving prediction models' accuracy has received less attention. This study proposes an integrated feature selection approach to predicting the Return on Assets (ROA) and Return on Equity (ROE) of manufacturing companies in Indonesia. This research involves data pre-processing, stationarity testing with Augmented Dickey-Fuller (ADF), and eliminating high multicollinearity variables using Variance Inflation Factor (VIF). In addition, feature selection was conducted through Feature Importance Ranking (FIR) and statistical regression p-value analysis to improve prediction efficiency. The results showed that feature selection in scenario 4 with the VIF integration approach and p-value significance successfully improved the Mean Squared Error (MSE) value of the Multilayer Perceptron (MLP) model by 96.11% and 90.72% in the Convolutional Neural Network (CNN) model compared to scenario 3 in predicting ROA. Random Forest is the best performing model in predicting ROE with an MSE value of 4.31828e-06. Meanwhile, the Light Gradient-Boosting Machine (LightGBM) model best predicts ROA with an MSE value of 8.75833e-07. Statistical regression models, especially the Random Effects Model (REM), have limitations in handling data with high multicollinearity. Therefore, this study confirms that integrating feature selection is important in improving prediction accuracy and that the LightGBM model is superior to conventional statistical approaches.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	deep learning, financial performance, machine learning, panel data, prediction model, analisa kinerja keuangan, data panel, deep learning, machine learning, model prediksi
Subjects:	T Technology > T Technology (General) > T174 Technological forecasting T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T58.62 Decision support systems
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User:	ANDY MOHAMMAD TEGUH
Date Deposited:	10 Feb 2025 00:36
Last Modified:	10 Feb 2025 00:36
URI:	http://repository.its.ac.id/id/eprint/118524

Actions (login required)

View Item