PARLIKA, Rizky and Isnanto, R. Rizal and Rahmat, Basuki (2025) TEKNIK DATASET TERPOLA SEBAGAI MODEL PREDIKSI UNTUK MENDUKUNG KEPUTUSAN INVESTASI CRYPTOCURRENCY MENGGUNAKAN PENDEKATAN DRM DAN CRISP-DM. Doctoral thesis, UNIVERSITAS DIPONEGORO.

	Text 01 Laporan Disertasi Rizky Parlika - Judul.pdf Download (271kB)
	Text COVER LENGKAP.pdf Restricted to Repository staff only Download (4MB)
	Text 16 Laporan Disertasi Rizky Parlika - BAB 1.pdf Download (500kB)
	Text 17 Laporan Disertasi Rizky Parlika - BAB 2.pdf Download (1MB)
	Text 18 Laporan Disertasi Rizky Parlika - BAB 3.pdf Restricted to Repository staff only Download (1MB)
	Text 19 Laporan Disertasi Rizky Parlika - BAB 4.pdf Restricted to Repository staff only Download (11MB)
	Text 20 Laporan Disertasi Rizky Parlika - BAB 5.pdf Restricted to Repository staff only Download (364kB)
	Text 21 Laporan Disertasi Rizky Parlika - Daftar Pustaka.pdf Download (669kB)
	Text 22 Laporan Disertasi Rizky Parlika - Lampiran - Lampiran.pdf Restricted to Repository staff only Download (807kB)

Abstract

Penelitian ini didasari banyaknya Banyak penelitian terdahulu secara langsung menggunakan dataset mentah sebagai data uji untuk menghasilkan prediksi Cryptocurrency, terutama Bitcoin tanpa diolah terlebih dahulu. Padahal pengolahan awal pada dataset mentah berpotensi menghasilkan Model Prediksi yang menjanjikan. Penelitian ini bertujuan mengembangkan Model Dataset Terpola untuk menentukan posisi beli dan jual aset Cryptocurrency, mengurangi risiko kerugian, serta mendeteksi awal hingga akhir fluktuasi, yang merupakan permasalahan yang ingin diselesaikan pada penelitian ini. Penelitian ini mengambil data awal dari pasar Cryptocurrency terbesar di Indonesia yakni Indodax. Data awal dikumpulkan dari periode Mei 2022 hingga Oktober 2024, menghasilkan lebih dari 3,3 juta data transaksi. Alat dan aplikasi yang digunakan mencakup bahasa pemrograman PHP, Python melalui Google Colaboratory, dan DBMS MySQL pada hosting serta server lokal. Model dataset terpola dibangun berdasarkan empat hipotesis utama, yang masing-masing berfokus pada mekanisme menentukan posisi beli, posisi jual, serta mendeteksi awal dan akhir fluktuasi harga. Keempat hipotesis yang diajukan terbukti benar, memberikan dasar yang kuat untuk pembuatan Teori Volatilitas dan Fluktuasi Cryptocurrency. Dalam membangun model dataset terpola, metodologi yang digunakan mengacu pada CRISP-DM (Cross-Industry Standard Process for Data Mining). Proses dimulai dengan pemahaman bisnis dan pemahaman data, diikuti dengan persiapan data, pembuatan model, evaluasi, dan implementasi hasil. Pengumpulan data dari Pasar Digital Indodax dilakukan melalui koneksi API. Dalam mengembangkan model, diterapkan metode Design Research Methodology (DRM) untuk merancang, menguji, dan mengevaluasi efektivitas model prediksi yang dihasilkan. Dataset Terpola berhasil menunjukkan titik beli dan jual yang berpotensi menguntungkan, serta mampu memetakan pola fluktuasi harga baik saat kenaikan maupun penurunan. Model Optimasi Hybrid LSTM (Long ShortTerm Memory) dengan Dataset Terpola yang di-resampel per 60 detik dengan data uji kurang lebih 1 juta data, menunjukkan akurasi prediksi yang lebih baik dari penelitian-penelitian sebelumnya dengan nilai MAPE 0.19%. LSTM dipilih karena menunjukkan hasil akurasi terbaik dari hasil perbandingan algoritma Statistical Learning, Machine Learning, dan Deep Learning. Hasil penelitian menunjukkan bahwa model prediksi yang dikembangkan dengan pendekatan DRM dan CRISP-DM efektif dalam memprediksi volatilitas dan fluktuasi harga Cryptocurrency, terutama Bitcoin. Model Prediksi Cryptocurrency baru ini berpotensi menghasilkan Keuntungan perdagangan dan mengurangi resiko kerugian. Validasi dilakukan dengan menghitung Potensi perolehan Capaian Return On Investment (ROI) dan Capaian Profit Maksimal dari historis dataset terpola sebelumnya, menunjukkan keuntungan yang konsisten.
Kata kunci: Model Dataset Terpola, Prediksi, Cryptocurrency, DRM, CRISP-DM

This research is based on many previous studies that directly use raw datasets as test data to generate cryptocurrency predictions, especially Bitcoin, without prior processing. In fact, initial processing of raw datasets has the potential to produce promising prediction models. This study aims to develop a patterned dataset model to determine the buy and sell positions of cryptocurrency assets, reduce the risk of loss, and detect the beginning and end of fluctuations, which are the problems to be solved in this study. This study takes initial data from the largest cryptocurrency market in Indonesia, namely Indodax. Initial data was collected from the period May 2022 to October 2024, resulting in more than 3.3 million transaction data. The tools and applications used include the PHP programming language, Python via Google Colaboratory, and MySQL DBMS on hosting and local servers. The patterned dataset model is built based on four main hypotheses, each of which focuses on the mechanism of determining buy positions, sell positions, and detecting the beginning and end of price fluctuations. The four hypotheses proposed were proven correct, providing a strong foundation for the creation of the Cryptocurrency Volatility and Fluctuation Theory. In building a patterned dataset model, the methodology used refers to CRISP-DM (Cross-Industry Standard Process for Data Mining). The process begins with business understanding and data understanding, followed by data preparation, model building, evaluation, and implementation of results. Data collection from the Indodax Digital Market is done through an API connection. In developing the model, the Design Research Methodology (DRM) method is applied to design, test, and evaluate the effectiveness of the resulting prediction model. The Patterned Dataset successfully shows potentially profitable buy and sell points, and is able to map price fluctuation patterns both during increases and decreases. The Hybrid LSTM (Long Short-Term Memory) Optimization Model with the Patterned Dataset resampled every 60 seconds with test data of approximately 1 million data, shows better prediction accuracy than previous studies with a MAPE value of 0.19%. LSTM was chosen because it showed the best accuracy results from the comparison of Statistical Learning, Machine Learning, and Deep Learning algorithms. The results of the study show that the prediction model developed with the DRM and CRISP-DM approaches is effective in predicting the volatility and price fluctuations of cryptocurrencies, especially Bitcoin. This new cryptocurrency prediction model has the potential to generate trading profits and reduce the risk of losses. Validation is carried out by calculating the potential for obtaining return on investment (ROI) and maximum profit achievement from the historical dataset of previous patterns, showing consistent profits.
Keywords: Patterned Dataset Model, Cryptocurrency, Prediction, DRM, CRISP- DM

Actions (login required)

View Item