Wahyu Sugianto, Herenda Sela Wismaya

Identifikasi Jenis Larutan Ionik Berdasarkan Data Spektroskopi Impedansi Listrik Menggunakan Unsupervised Machine learning

Introduction

Identifikasi jenis larutan ionik berdasarkan data spektroskopi impedansi listrik menggunakan unsupervised machine learning. Penentuan jenis larutan ionik via Spektroskopi Impedansi Listrik & Unsupervised Machine Learning (PCA, K-Means). Solusi cepat, akurat 88.9% untuk biomedis.

Abstract

Identifikasi jenis larutan ionik merupakan salah satu bagian fundamental dalam bidang biomedis. Spektroskopi Impedansi Listrik (SIL) merupakan sebuah metode yang terbukti mampu mengidentifikasi proses elektokimia yang unik untuk setiap larutan, namun interpretasi hasil pengukuran pada SIL terbilang cukup rumit. Disisi lain, metode identifikasi secara cepat, akurat, dan tanpa reagen menjadi kebutuhan penting dalam bidang biomedis dan lingkungan. Penelitian ini bertujuan untuk mengevaluasi potensi penggunaan Unsupervised Machine learning, khususnya metode Principal component analysis (PCA) dan K-Means untuk menginterpretasikan hasil pengukuran spektroskopi impedansi listrik (SIL) sehingga mampu membedakan jenis larutan ionik NaCl, Ringer Laktat (RL), dan Simulated Body Fluid (SBF). Pengukuran impedansi dilakukan pada rentang frekuensi 1 Hz hingga 1 MHz menggunakan dua elektroda tanpa modifikasi permukaan. Selanjutnya, hasil pengukuran impedani berupa dan dianalisis untuk mengekstrak data. Proses analisis meliputi pra-pemrosesan data, reduksi dimensi menggunakan Principal component analysis (PCA), dan klasterisasi dengan algoritma K-Means. PCA mampu merangkum informasi sebesar 75.8% dalam dua komponen utama, dan hasil klasterisasi menunjukkan pemisahan yang cukup jelas antara ketiga jenis larutan dengan nilai accuracy sebesar 88.9%, Adjusted Rand Index (ARI) sebesar 0.6865, serta silhouette score sebesar 0.4473. Hasil ini menunjukkan bahwa penggunaan kombinasi algortima PCA dan K-Means memiliki potensi untuk interpretasi hasil SIL yang tidak memerlukan data berlabel (unsupervised) dengan performa yang kompetitif jika dibandingkan dengan metode supervised, sehingga lebih fleksibel untuk aplikasi dimana pelabelan data sulit dilakukan.

Review

This paper addresses the fundamental challenge of identifying ionic solutions, a critical task in biomedical applications, by proposing an innovative approach combining Electrical Impedance Spectroscopy (EIS) with unsupervised machine learning. Recognizing the inherent complexity in interpreting raw EIS data, the authors successfully demonstrate the potential of Principal Component Analysis (PCA) for dimensionality reduction and K-Means for clustering to differentiate between three distinct ionic solutions: NaCl, Ringer Lactate, and Simulated Body Fluid. The study details measurements taken across a broad frequency range (1 Hz to 1 MHz), leading to an interpretable model. The results showcase promising performance, with the combined PCA and K-Means algorithm achieving an accuracy of 88.9%, an Adjusted Rand Index (ARI) of 0.6865, and a silhouette score of 0.4473, effectively distinguishing the solution types without requiring labeled training data. The primary strength of this research lies in its novel and practical application of unsupervised machine learning to simplify the interpretation of complex EIS data. The ability to achieve robust identification of ionic solutions without the need for pre-labeled datasets offers significant advantages, particularly in biomedical and environmental settings where data annotation can be laborious and expensive. The successful integration of PCA to distill meaningful information (75.8% in two components) from high-dimensional impedance spectra, followed by K-Means clustering, provides a clear and efficient methodology. This unsupervised approach not only demonstrates competitive performance compared to supervised methods but also enhances the flexibility and applicability of EIS for real-world scenarios where diverse and unpredictable samples are encountered. While the study presents compelling results, there are avenues for further exploration and improvement. The current scope is limited to three well-defined ionic solutions, and future work could investigate the scalability and robustness of this approach when applied to a broader spectrum of ionic mixtures or more complex biological fluids with varying compositions and interferences. Furthermore, although the unsupervised approach is valuable, a direct comparison with state-of-the-art supervised learning methods on the exact same dataset could provide a more rigorous benchmark for the term "competitive performance." Future research could also explore the impact of different electrode designs or surface modifications on measurement sensitivity and specificity. Finally, exploring the potential for real-time analysis and miniaturization of the system would be crucial steps towards clinical or field deployment.

Full Text

Comments

You need to be logged in to post a comment.