Deteksi Anomali Pasar Saham Berbasis Algoritma Mapper Dengan HDBSCAN

Yumna, Mutsaqoful Izah (2025) Deteksi Anomali Pasar Saham Berbasis Algoritma Mapper Dengan HDBSCAN. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5002211106-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Deteksi anomali pada pasar saham merupakan aspek krusial dalam memahami dinamika pasar yang kompleks dan bersifat non-linier. Metode konvensional seperti Bollinger Bands dan One-Class Support Vector Machine (OCSVM) memiliki keterbatasan dalam menangkap pola struktural yang tersembunyi dalam data berdimensi tinggi. Oleh karena itu, penelitian ini bertujuan untuk mengimplementasikan dan mengevaluasi efektivitas kombinasi algoritma Mapper dan HDBSCAN sebagai pendekatan Topological Data Analysis (TDA) dalam mendeteksi anomali pada data harga penutupan saham indeks LQ45 selama periode 1 Januari 2020 hingga 30 April 2025. Hasil penelitian menunjukkan bahwa Algoritma Mapper diduga mengidentifikasi anomali yang berkorespondensi dengan pola harga pada fenomena Walking the Bands pada kondisi overbought dan oversold yang mengidentifikasi sebuah tren berlanjut. Evaluasi menunjukkan bahwa hasil One-Class SVM memiliki performa terbaik dengan akurasi 91% dan recall 52%, menjadikannya paling sensitif dalam mengidentifikasi anomali. Bollinger Bands menunjukkan performa yang cukup baik dengan akurasi 88% dan recall 43%. Sebaliknya, Algoritma Mapper memiliki keterbatasan, meskipun recall-nya tertinggi 65%, presisinya sangat rendah 16%, yang mengindikasikan banyak kesalahan deteksi. Kinerja ini menunjukkan Mapper gagal menyaingi performa dari kedua metode lainnya. Selain itu, Algoritma Mapper juga memiliki efisiensi komputasi yang lebih rendah.
==================================================================================================================================
Anomaly detection in the stock market is a crucial aspect of understanding the complex and non-linear dynamics of financial systems. Conventional methods such as Bollinger Bands and the One-Class Support Vector Machine (OCSVM) face limitations in capturing hidden structural patterns within high-dimensional data. Therefore, this study aims to implement and evaluate the effectiveness of a combination of the Mapper algorithm and HDBSCAN as a Topological Data Analysis (TDA) approach for detecting anomalies in the closing price data of the LQ45 stock index over the period from January 1, 2020, to April 30, 2025. The results suggest that the Mapper algorithm potentially identifies anomalies corresponding to price patterns associated with the Walking the Bands phenomenon under overbought and oversold conditions, which often indicate a continuing trend. Evaluation results show that the One-Class SVM achieves the best performance, with an accuracy of 91% and a recall of 52%, making it the most sensitive method for anomaly detection. Bollinger Bands also demonstrate reasonably strong performance, with an accuracy of 88% and a recall of 43%. In contrast, the Mapper algorithm shows notable limitations. Although it yields the highest recall (65%), its precision is significantly low (16%), indicating a high rate of false positives. This performance suggests that Mapper falls short of the other two methods. Moreover, the Mapper algorithm is less computationally efficient compared to its counterparts.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Anomali Pasar Saham, Algoritma Mapper, HDBSCAN, Topological Data Analysis, Stock Market Anomalies, Mapper Algorithm, HDBSCAN, Topological Data Analysis
Subjects:	Q Science Q Science > QA Mathematics Q Science > QA Mathematics > QA166 Graph theory Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry)
Divisions:	Faculty of Mathematics and Science > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Mutsaqoful Izah Yumna
Date Deposited:	01 Aug 2025 05:48
Last Modified:	01 Aug 2025 05:48
URI:	http://repository.its.ac.id/id/eprint/125952

Actions (login required)

View Item