Imputasi Missing Value pada Data Deret Waktu Tekanan Air dengan Model Pembelajaran Mesin Berbasis Atensi

Susanto, Tri (2025) Imputasi Missing Value pada Data Deret Waktu Tekanan Air dengan Model Pembelajaran Mesin Berbasis Atensi. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022231060-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (15MB) | Request a copy

Abstract

Missing value pada data deret waktu menjadi tantangan signifikan dalam sistem Internet of Things (IoT), khususnya pada aplikasi monitoring tekanan air yang bergantung pada kontinuitas dan akurasi data. Penelitian ini mengusulkan pendekatan imputasi berbasis pembelajaran mesin dengan arsitektur Bidirectional Long Short-Term Memory (Bi-LSTM) yang dipadukan dengan mekanisme Multi-Head Attention, guna meningkatkan ketepatan estimasi missing value. Model yang dikembangkan tidak hanya menangkap dinamika temporal dua arah secara simultan, tetapi juga mampu menyesuaikan fokus perhatian terhadap informasi yang paling relevan, termasuk pada data dari berbagai perangkat sensor yang memiliki karakteristik berbeda. Penelitian ini mengevaluasi performa model terhadap beberapa strategi Fine-Tuning dan membandingkannya dengan metode konvensional seperti Mean Imputation, Median Imputation, Regresi Linear, serta LSTM dan Bi-LSTM tanpa atensi. Uji eksperimental dilakukan dengan variasi tingkat sparsitas dan skenario missing data (MCAR dan MAR), serta melibatkan data tekanan air riil yang dikumpulkan dari sistem IoT Perumda Air Minum Surya Sembada Kota Surabaya. Hasil evaluasi menunjukkan bahwa model Bi-LSTM dengan Multi-Head Attention versi kedua (V2) secara konsisten menghasilkan tingkat error yang lebih rendah (MAE, MSE, RMSE, dan MAPE) dibandingkan baseline lainnya, dan lebih konsisten di berbagai kondisi perangkat. Dengan demikian, pendekatan yang diusulkan menawarkan solusi yang adaptif dan presisi tinggi dalam menghadapi permasalahan missing value pada data deret waktu IoT, sekaligus memperluas potensi penerapan model atensi dalam pemrosesan data sensor untuk mendukung pengambilan keputusan berbasis data di sektor utilitas.

=========================================================================================================================================
Missing values in time series data present a significant challenge in Internet of Things (IoT) systems, particularly in water pressure monitoring applications where data continuity and accuracy are critical. This study proposes a machine learning-based imputation approach using a Bidirectional Long Short-Term Memory (Bi-LSTM) architecture enhanced with a Multi-Head Attention mechanism to improve the accuracy of missing value estimation. The proposed model not only captures bidirectional temporal dynamics simultaneously but also adaptively focuses attention on the most relevant information, even across heterogeneous sensor devices with varying data characteristics. The model’s performance is evaluated under several fine-tuning strategies and benchmarked against conventional imputation methods such as Mean Imputation, Median Imputation, Linear Regression, as well as LSTM and Bi-LSTM without attention. Experimental testing was conducted using varying missing rates and missingness mechanisms (MCAR and MAR), applied to real-world water pressure data collected from the IoT infrastructure of Perumda Air Minum Surya Sembada Kota Surabaya. Results demonstrate that the Bi-LSTM with Multi-Head Attention version two (V2) consistently achieves lower error rates (MAE, MSE, RMSE, and MAPE) compared to baseline methods, and performs robustly across different sensor devices. In conclusion, the proposed approach offers an adaptive and high-precision solution to missing value problems in time series IoT data, while also extending the applicability of attention-based models for sensor data processing to support data-driven decision-making in utility management.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	imputasi, missing value, deret waktu, pembelajaran mesin, Bi-LSTM, atensi, IoT, tekanan air, missing value, imputation, time series, machine learning, attention, IoT, water pressure
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Tri Susanto -
Date Deposited:	22 Jul 2025 05:51
Last Modified:	22 Jul 2025 05:51
URI:	http://repository.its.ac.id/id/eprint/120500

Actions (login required)

View Item