Pendekatan Model Agnostik Untuk Menginterpretasikan Model Machine Learning Pada Kasus Klasifikasi (Studi Kasus Prediksi Financial Distress Perusahaan Sektor Industri Di Indonesia)

Alfitri, Ratnanda Gita (2021) Pendekatan Model Agnostik Untuk Menginterpretasikan Model Machine Learning Pada Kasus Klasifikasi (Studi Kasus Prediksi Financial Distress Perusahaan Sektor Industri Di Indonesia). Undergraduate thesis, Institut Teknologi Sepuluh Nopember.

Text 06211740000086-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2023. Download (8MB) | Request a copy

Abstract

Kompleksitas model machine learning menjadikan model ini sulit diinterpretasi sehingga disebut model black box. Padahal, kemampuan untuk menafsirkan output model prediksi sangat penting karena dapat mempengaruhi keyakinan pengguna terhadap hasil prediksi. Oleh karena itu, pendekatan model agnostik perlu dilakukan. Penelitian ini bertujuan untuk menginterpretasikan model machine learning dalam mengklasifikan kondisi keuangan perusahaan di sektor industri. Model machine learning yang digunakan ialah GEVR, XGB, dan NN. Kinerja NN lebih unggul pada proses training, namun tidak lebih baik di data testing. Model agnostik yang digunakan adalah Partial Dependence Plot, Feature Interaction, Permutation Feature Importance, Global dan Local Surrogate, dan Shapley Value. Feature selection menyeleksi variabel Current Ratio dan Return of Equity. Penanganan outlier dengan winsorization 2,5% menghasilkan perbedaan hasil pada model agnotik. Pola pada Partial Dependence Plot semakin terlihat dalam range yang lebih sempit. Tidak terdapat perbedaan error setelah permutasi variabel pada GEVR dan XGB tidak muncul. Prediksi secara lokal dengan winsorization lebih sesuai dengan kondisi aktual.
=====================================================================================================
The complexity of machine learning models makes this model difficult to interpret, so it is called a black-box model. The ability to interpret the output of the predictive model is very important because it can affect the user's belief in the prediction results. Therefore, a model agnostic approach needs to be carried out. This study aims to interpret machine learning models in classifying the financial condition of companies in the industrial sector. The machine learning models used are GEVR, XGB, and NN. NN is superior to the training process, but not better at data testing. The agnostic models used are Partial Dependence Plot, Feature Interaction, Permutation Feature Importance, Global and Local Surrogate, and Shapley Value. Feature selection remove Current Ratio and Return of Equity variables. Handling outliers with 2.5% winsorization resulted in different results in the agnostic model. The pattern in the Partial Dependence Plot is increasingly visible in a narrower range. There are no difference of error after permutation on GEVR and XGB. Local prediction with winsorization is more close to actual condition.

Item Type:	Thesis (Undergraduate)
Uncontrolled Keywords:	Financial Distress, Interpretable Machine Learning, Klasifikasi, Model Agnostik, Prediksi, Classification, Model-Agnostic, Prediction
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Ratnanda Gita Alfitri
Date Deposited:	03 Sep 2021 03:31
Last Modified:	03 Sep 2021 03:31
URI:	http://repository.its.ac.id/id/eprint/91618

Actions (login required)

View Item