cover
Contact Name
Amalia Eka Rakhmania
Contact Email
amaliaeka.rakhmania@polinema.ac.id
Phone
+6282134134594
Journal Mail Official
eltek@polinema.ac.id
Editorial Address
Pusat Penelitian dan Pengabdian Masyarakat Politeknik Negeri Malang Graha Polinema Lantai 3 Jl. Sukarno Hatta No. 9, Malang, Indonesia, 65141
Location
Kota malang,
Jawa timur
INDONESIA
Jurnal ELTEK
ISSN : 16934024     EISSN : 23550740     DOI : doi.org/10.33795/eltek
Jurnal ELTEK merupakan media yang diterbitkan oleh Politeknik Negeri Malang sebagai sarana diseminasi dan publikasi artikel hasil penelitian dan artikel konseptual yang dilakukan oleh para peneliti, akademisi, praktisi dan industri di bidang Teknik Elektro. Artikel yang diajukan untuk diterbitkan pada jurnal ELTEK merupakan naskah asli dan belum pernah dipublikasikan secara tertulis pada majalah atau jurnal ilmiah dimanapun Jurnal ELTEK memuat artikel hasil penelitian dan konseptual dalam cakupan bidang ilmu Teknik Elektro yang meliputi: Ketenagalistrikan, Teknik Kendali, Kecerdasan Buatan, Teknik Komputer, Teknik Telekomunikasi, Teknik informatika, Teknik Biomedik, Elektronika Daya, Energi Terbarukan, Sistem Embedded, Jaringan Komputer, Sistem Operasi, Manajemen Data dan Sistem informasi, dan lain-lain. Jurnal ELTEK terbit dua kali dalam setahun masing-masing pada bulan April dan Oktober. Redaksi mengundang penulis dan peneliti untuk menyumbangkan artikel penelitian atau artikel konseptual kepada Editor Jurnal ELTEK. Editor menentukan apakah suatu artikel akan dimuat atau tidak dan memiliki hak untuk mengubah atau mengoreksi teks sejauh tidak mengganggu maksud atau kontennya.
Articles 86 Documents
Implementasi sistem penyemprotan herbisida cerdas berbasis wireless sensor network untuk pertanian presisi Nailul Muna; Norma Ningsih; Nanang Syahroni; Ekananda Sulistyo Putra; Muhammad Naufal Asyam Muflih; Muhamad Nur Afriza Farkhan
JURNAL ELTEK Vol. 24 No. 1 (2026): April 2026
Publisher : Politeknik Negeri Malang

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33795/eltek.v24i1.9865

Abstract

Penelitian ini mengembangkan sistem penyemprotan herbisida otomatis berbasis Internet of Things (IoT) dan Wireless Sensor Network (WSN) untuk mendukung pertanian presisi. Sistem terdiri atas dua node penyemprotan yang dilengkapi Raspberry Pi 5, ESP32, IP Camera, sensor ultrasonik, sensor waterflow, relay, dan pompa penyemprot. Raspberry Pi digunakan sebagai edge processing unit, sedangkan ESP32 berfungsi sebagai pengendali sensor dan aktuator. Sistem menggunakan SSD-MobileNet-V2 FPNLite untuk mendeteksi keberadaan gulma sebagai mekanisme pendukung pengambilan keputusan penyemprotan otomatis. Seluruh data monitoring dikirimkan ke Firebase dan divisualisasikan melalui website secara real-time. Hasil pengujian menunjukkan bahwa rata-rata akurasi sensor ultrasonik setelah kalibrasi mencapai 97,53% pada Node 1 dan 97,59% pada Node 2, sedangkan sensor waterflow mencapai 95,94% pada Node 1 dan 97,43% pada Node 2. Pengujian komunikasi data menunjukkan rata-rata latency sebesar 1,9 detik. Selain itu, seluruh perangkat keras dan website monitoring berhasil beroperasi dengan baik selama pengujian sistem. Hasil penelitian menunjukkan bahwa sistem mampu mengintegrasikan proses monitoring, komunikasi data, dan penyemprotan herbisida secara otomatis.   ABSTRACT An automatic herbicide spraying hardware system based on the Internet of Things (IoT) and Wireless Sensor Network (WSN) was developed to support precision agriculture. The system consists of two spraying nodes equipped with Raspberry Pi 5, ESP32, IP Camera, ultrasonic sensor, waterflow sensor, relay, and spraying pump. Raspberry Pi functions as the edge processing unit, while ESP32 serves as the sensor and actuator controller. The system utilizes SSD-MobileNet-V2 FPNLite to detect weeds as a supporting mechanism for automatic spraying decision-making. All monitoring data are transmitted to Firebase and visualized through a real-time monitoring website. Experimental results show that the average error of the calibrated ultrasonic sensor reached 2.47% on Node 1 and 2.41% on Node 2, while the waterflow sensor achieved 4.06% on Node 1 and 2.57% on Node 2. Data communication testing showed an average latency of 1.9 seconds. In addition, all hardware components and the monitoring website operated properly during system testing. The experimental results demonstrate that the developed system is capable of integrating monitoring, data communication, and automatic herbicide spraying processes.
Implementation of user-centered design and white-box testing for a laboratory equipment loan application Mohamad Dimyati Ayatullah; I Wayan Suardinata; Mega Devita Sari; Lukman Hakim; Aracely Amanda Gusta
JURNAL ELTEK Vol. 24 No. 2 (2026): October 2026
Publisher : Politeknik Negeri Malang

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33795/eltek.v24i2.5543

Abstract

The laboratory equipment loan process in the Department of Business and Informatics at Politeknik Negeri Banyuwangi was previously managed using paper-based forms, which resulted in limited administrative efficiency and the absence of real-time loan status monitoring. This study aims to develop a web-based laboratory equipment loan application using the User-Centered Design (UCD) approach and to evaluate the reliability of the proposed system workflow through White Box Testing. The UCD process involved identifying user requirements from four user roles, resulting in 14 functional requirements that were fully implemented in the developed system. A task-flow comparison between the existing manual process and the proposed system showed that the minimum number of steps required for three representative loan activities was reduced from eight steps to four steps, indicating a 50% reduction in process complexity. Furthermore, White Box Testing on the student and lecturer loan modules obtained a cyclomatic complexity value of V(G)=2, with all four test scenarios successfully executed. The results demonstrate that the developed system provides complete functional requirement coverage and simplifies the laboratory equipment loan workflow. However, this study does not evaluate processing time improvement or user satisfaction through standardized usability instruments; therefore, claims regarding statistical performance improvement or user acceptance are beyond the scope of this work.
Perancangan UI/UX aplikasi notulensi rapat berbasis web dengan metode design thinking di Politeknik Negeri Banyuwangi Eka Novita Sari; Alfin Hidayat; Ajeng Fuji; Putri Dewi Hendista; Aghisni Zuhruffi
JURNAL ELTEK Vol. 24 No. 2 (2026): October 2026
Publisher : Politeknik Negeri Malang

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33795/eltek.v24i2.6178

Abstract

Rapat merupakan agenda penting yang rutin dilaksanakan oleh civitas akademik di Politeknik Negeri Banyuwangi. Kegiatan ini dilakukan untuk melakukan perundingan dan membahas guna menyelesaikan sebuah permasalahan yang berkaitan dengan kepentingan kampus. Didalam sebuah rapat tentunya tidak lepas dari sebuah agenda yang biasa dikenal dengan notulensi, namun yang menjadi sebuah permasalahan pada Pusat Penjaminan Mutu dan Pengembangan Pembelajaran di Politeknik Negeri Banyuwangi adalah notulensi yang masih dilakukan secara manual. Dari permasalahan tersebut kemudian dikembangkan sebuah solusi yaitu Pengembangan UI/UX aplikasi notulensi berbasis Web dengan menggunakan metode design thinking. Pada hasil dari pengembangan yang sudah dilakukan dengan menggunakan metode System Usability Scale mendapatkan hasil dengan rata-rata nilai sebanyak 85.23 yang berarti sistem yang sudah dibangun sejauh ini ini berada pada kategori Excellent dan dianggap sudah bisa mengatasi permasalahann yang ada di Pusat Penjaminan Mutu dan Pengembangan Pembelajaran yang ada di Politeknik Negeri Banyuwangi. Abstract Meetings are an important agenda routinely conducted by the academic community at the Banyuwangi State Polytechnic. These activities are held to discuss and address issues related to the interests of the campus. Meetings inevitably involve the process of minute-taking, but a significant issue at the Center for Quality Assurance and Learning Development at Banyuwangi State Polytechnic is that the minutes are still taken manually. To address this issue, a solution was developed: the design and development of a web-based minute-taking application using the design thinking method. The results of this development, evaluated using the System Usability Scale (SUS) method, yielded an average score of 85.23, indicating that the system falls within the "Excellent" category and is considered capable of resolving the issues at the Center for Quality Assurance and Learning Development at Banyuwangi State Polytechnic
A color-based particle filter optimized using firefly and black hole algorithms for moving object tracking Devira Anggi Maharani; Lenni Yulianti; Carmadi Machbub; Pranoto Hidaya Rusmin; Suci Aulia; Rizki Dwi Romadhona
JURNAL ELTEK Vol. 24 No. 2 (2026): October 2026
Publisher : Politeknik Negeri Malang

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33795/eltek.v24i2.9879

Abstract

Many researchers, particularly in the field of computer vision, have increasingly focused on moving object tracking. However, accurately tracking a moving object remains challenging, especially under conditions such as occlusion, illumination changes, abrupt motion, and variations in object posture. Although recent developments have enhanced the performance of visual object tracking methods, the task remains challenging under real-world conditions. Particle Filter (PF), a Monte Carlo-based estimation framework, has been extensively applied to visual object tracking due to its ability to handle nonlinear and non-Gaussian systems. However, conventional PF methods still suffer from particle degeneracy and sample impoverishment, particularly after repeated resampling processes. Although resampling can reduce degeneracy by removing particles with insignificant weights, it may also decrease particle diversity and degrade tracking performance. To address these limitations, this study introduces a Black Hole Optimization Particle Filter (BHOPF), in which the Black Hole Optimization algorithm is incorporated into the PF resampling process. The proposed approach updates particles through the Black Hole optimization mechanism, offering a simple computational structure with minimal parameter tuning. The integration of Black Hole Optimization is expected to improve particle diversity, increase the effective number of particles, reduce RMSE, and enhance the robustness of PF-based visual object tracking. The experimental results show that, using 30 particles and 5 iterations, BHOPF increased the effective number of particles to 26. In comparison, the Firefly Optimization Particle Filter (FAOPF) achieved 23 effective particles, while the conventional Particle Filter achieved only 16 effective particles. The proposed system achieved real-time operation through multithreading and pipeline-based processing, with a processing speed of up to 30 FPS.
Ekstraksi fitur tekstur kerusakan situs warisan budaya pascabencana menggunakan deep features efficientnet-B7 Muhammad Hatif; Sri Harini; Ririen Kusumawati; Zainal Abidin; Muhammad Ainul Yaqin; Agung Teguh Wibowo Almais
JURNAL ELTEK Vol. 24 No. 2 (2026): October 2026
Publisher : Politeknik Negeri Malang

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33795/eltek.v24i2.9886

Abstract

Situs warisan budaya sangat rentan mengalami kerusakan akibat bencana alam, sehingga memerlukan metode inspeksi yang cepat, akurat, dan objektif untuk mendukung upaya konservasi. Penelitian ini mengusulkan pendekatan ekstraksi dan karakterisasi fitur tekstur otomatis melalui pemanfaatan fitur mendalam (deep feature extraction) menggunakan arsitektur EfficientNet-B7 sebagai landasan untuk deteksi kerusakan. Metode ini memanfaatkan model pre-trained untuk mengekstraksi peta fitur dari lapisan awal block2a_expand_activation, yang peka terhadap detail tekstural seperti tepi, retak, dan variasi permukaan. Dari 144 saluran fitur yang dihasilkan, dihitung tiga deskriptor statistik sederhana per saluran—Kontras (standar deviasi), Energi (rata-rata kuadrat aktivasi), dan Homogenitas (turunan kontras)—sebagai deskriptor tekstur kuantitatif (bukan fitur GLCM klasik). Sebagai studi awal (proof of concept), pendekatan ini diujicobakan pada satu citra kerusakan plafon masjid pascabencana dan berhasil mengkarakterisasi variasi tekstur antar-saluran, menghasilkan 432 fitur numerik (3 statistik × 144 saluran). Secara konseptual, dibandingkan metode tradisional seperti Gray Level Co-occurrence Matrix (GLCM), pendekatan ini berpotensi lebih unggul dalam hal otomasi, kedalaman fitur, dan skalabilitas karena tidak memerlukan parameterisasi manual, meskipun perbandingan empiris langsung masih perlu dilakukan. Meskipun membutuhkan sumber daya komputasi yang signifikan, metode yang diusulkan menawarkan solusi yang efisien dan andal untuk analisis kerusakan otomatis, yang dapat mendukung prioritisasi dan pemulihan situs warisan budaya pascabencana. Abstract Cultural heritage sites are highly vulnerable to damage from natural disasters, necessitating fast, accurate, and objective inspection methods to support conservation efforts. This study proposes an automated approach for texture feature extraction and characterization through deep feature extraction using the EfficientNet-B7 architecture, serving as a foundation for damage detection. This method utilizes a pre-trained model to extract feature maps from the initial block2a_expand_activation layer, which is sensitive to textural details such as edges, cracks, and surface variations. From the 144 feature channels generated, three simple per-channel statistical descriptors were computed—Contrast (standard deviation), Energy (mean squared activation), and Homogeneity (derived from contrast)—as quantitative texture descriptors. As a proof of concept, the approach was tested on a single mosque ceiling damage image and successfully characterized inter-channel texture variations, producing 432 numerical features (3 statistics × 144 channels). Conceptually, compared to traditional methods such as Gray Level Co-occurrence Matrix (GLCM), this approach is potentially superior in terms of automation, feature depth, and scalability, as it does not require manual parameterization, although a direct empirical comparison remains to be conducted. Although requiring significant computational resources, the proposed method offers an efficient and reliable solution for automated damage analysis, which can support the prioritization and restoration of cultural heritage sites after disasters.
Pelabelan tingkat guncangan gempa bumi menggunakan clustering berbasis principal component analysis Agung Teguh Wibowo; Renata Amalia; Supriyono
JURNAL ELTEK Vol. 24 No. 2 (2026): October 2026
Publisher : Politeknik Negeri Malang

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33795/eltek.v24i2.9955

Abstract

Indonesia memiliki tingkat kerentanan tinggi terhadap bencana alam, khususnya gempa bumi. Perbedaan tingkat guncangan gempa memerlukan upaya mitigasi berbasis teknologi untuk mendukung analisis data gempa. Penelitian ini bertujuan mengelompokkan data gempa menggunakan metode Principal Component Analysis berbasis clustering. Data yang digunakan berupa data gempa harian tahun 2024 dari Badan Meteorologi, Klimatologi, dan Geofisika Stasiun Geofisika Kelas III Malang. Metode ini digunakan untuk mengekstraksi komponen utama yang merepresentasikan variasi dominan data. Hasil analisis menunjukkan bahwa komponen utama pertama mampu menjelaskan 34,1% variasi data dan didominasi oleh variabel longitude serta magnitudo. Nilai komponen utama pertama digunakan sebagai dasar proses clustering untuk mengidentifikasi pola gempa berdasarkan kemiripan karakteristik. Pengujian Calinski-Harabasz Index menghasilkan nilai sebesar 5,626,0269. Berdasarkan hasil tersebut, data gempa dikelompokkan menjadi kategori dirasakan dan tidak dirasakan mengacu pada Skala Intensitas Gempabumi Badan Meteorologi, Klimatologi, dan Geofisika. Hasil penelitian menunjukkan bahwa metode tersebut mampu mengurangi kompleksitas data dan membantu pengelompokan tingkat guncangan gempa secara efektif dan efisien. ABSTRACT Indonesia is highly vulnerable to natural disasters, particularly earthquakes. This study aims to classify earthquake data using a clustering-based Principal Component Analysis method. The dataset consists of daily earthquake records in 2024 from the Badan Meteorologi, Klimatologi, dan Geofisika Geophysics Station Class III Malang. Principal component analysis was used to extract the main components representing dominant data variations. The results showed that the first principal component explained 34,1% of the variance and was dominated by longitude and magnitude variables. The first principal component scores were then used for clustering based on earthquake characteristics similarity. The Calinski-Harabasz Index produced a value of 5,626.0269. Based on the clustering results, earthquakes were categorized into felt and unfelt groups according to the Earthquake Intensity Scale of Badan Meteorologi, Klimatologi, dan Geofisika. The findings demonstrate that the method effectively reduces data complexity and supports efficient earthquake shaking classification.
Edge-based multi-object PPE compliance monitoring: a critical evaluation of YOLOv8n structural reliability on resource-constrained platforms Wahyu Nur Hidayat; Moechammad Sarosa; Sapto Wibowo; Erdiyan Ariyanda
JURNAL ELTEK Vol. 24 No. 2 (2026): October 2026
Publisher : Politeknik Negeri Malang

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33795/eltek.v24i2.10286

Abstract

Occupational Safety and Health (OSH) requires continuous monitoring of Personal Protective Equipment (PPE) compliance, while manual supervision is limited by coverage, continuity, and human observation. This study evaluates an edge-based PPE monitoring prototype using YOLOv8n on a Raspberry Pi 4 for detecting six PPE classes: helmet, safety glasses, mask, safety vest, gloves, and safety shoes. The system uses a 2K QHD autofocus USB camera mounted on a static tripod and integrates object detection with a tiered SOP decision module and physical warning output. Model performance was evaluated using precision, recall, F1-score, Average Precision (AP), mean Average Precision (mAP), FPS, and inference time. Field evaluation included six initial test frames, distance tests at 3-5 m with frontal and side orientations, additional testing at 2 m, a dedicated safety-glasses test at approximately 1 m, and a qualitative indoor test. The model achieved an overall mAP@0.5 of 0.737. Helmet, safety vest, and safety shoes showed the strongest AP@0.5 values, whereas smaller or less visually distinct PPE classes remained more difficult to detect. Edge deployment achieved 8.4-11.6 FPS with inference times of 86.1-119.3 ms. Across the six initial test frames, five SOP decisions were correct, resulting in 83.3% decision accuracy and a 16.7% false alarm rate. The results demonstrate the feasibility of limited real-time PPE monitoring on Raspberry Pi 4 while showing that reliability remains sensitive to object size, distance, viewpoint, and occlusion.
Annealing effects on single-phase E-I transformer performance under nonlinear loads Slamet Nurhadi; Chandra Wiharya; Satria Luthfi Hermawan; Priya Surya Harijanto; Bakti Indra Kurniawan; Divac Nabiel Akbar
JURNAL ELTEK Vol. 24 No. 2 (2026): October 2026
Publisher : Politeknik Negeri Malang

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33795/eltek.v24i2.10746

Abstract

Single-phase E-I transformers are widely used in power-conversion applications, including grid-connected flyback micro-inverters for solar photovoltaic systems, yet their performance often deteriorates due to high core loss and harmonic distortion generated by nonlinear loads. This study evaluated the effect of the annealing process on core loss, voltage regulation, energy efficiency, and voltage/current harmonic distortion of a single-phase E-I transformer under nonlinear loading. Three transformer samples were tested under no-load and nonlinear-load conditions (rectifier and inverter) before and after annealing at 180, 350, and 450 °C for 60 minutes, followed by harmonic-distortion measurement using a power logger on a flexible conveyor system and a single-phase induction motor. Annealing at 350 °C produced the most significant core-loss reduction, from 150.3 W to 110.6 W, while linear-load efficiency reached 97-98% and voltage regulation remained stable near 98%. Harmonic measurements showed voltage total harmonic distortion (THDv) of 5.8% on the flexible conveyor system and 5.6% on the induction motor without a variable speed drive, both below the 8% IEEE 519 limit, while individual current harmonic distortion (IHDi), current total harmonic distortion (THDi), and total demand distortion (TDD) also remained within the standard's safe margins. These findings indicate that medium-temperature annealing can improve the magnetic performance of the transformer core while keeping power quality within standard limits under nonlinear loading. The practical implication is that this annealing guideline can be adopted by small and medium-scale transformer manufacturers to improve energy efficiency without costly equipment investment.
Pengenalan wajah berbasis convolutional neural network pada smart parcel box untuk keamanan paket di kos Siti Farida; Putri Elfa Mas'udia; Sri Wahyuni Dali
JURNAL ELTEK Vol. 24 No. 2 (2026): October 2026
Publisher : Politeknik Negeri Malang

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33795/eltek.v24i2.10806

Abstract

Maraknya belanja online di kalangan mahasiswa yang tinggal di kos menimbulkan permasalahan keamanan paket, terutama risiko kehilangan paket yang sulit ditelusuri akibat keterbatasan sistem keamanan konvensional. Penelitian ini bertujuan merancang dan mengimplementasikan Smart Parcel Box berbasis pengenalan wajah menggunakan Convolutional Neural Netra sebagai solusi sistem keamanan paket di area kos. Sistem mengintegrasikan Raspberry Pi 3B+ sebagai unit pemrosesan utama, ESP32 sebagai mikrokontroler, sensor GM67 untuk pemindaian nomor resi, sensor ultrasonik HC-SR04 untuk pemantauan kapasitas kotak, serta solenoid door lock sebagai aktuator pengunci pintu. Pengenalan wajah menggunakan multi-task cascaded convolutional network untuk deteksi wajah dan model FaceNet berbasis TensorFlow Lite untuk pengenalan wajah, dengan Cosine Similarity sebagai metode pencocokan pada nilai threshold 0,75. Sistem dilengkapi aplikasi mobile Android yang terintegrasi dengan Firebase sebagai database real-time untuk manajemen data pengguna, resi paket, dan riwayat pengambilan. Hasil pengujian menunjukkan sistem face recognition mencapai akurasi 85,8%. Sensor GM67 mencapai tingkat keberhasilan baca barcode 92,3% serta jarak optimal pembacaan 10–40 cm, sedangkan sensor ultrasonik menghasilkan akurasi rata-rata 99,49% dengan error rata-rata 0,51%. ABSTRACT The surge in online shopping among boarding house residents has raised concerns over package security, particularly theft that is difficult to trace due to limitations of conventional security systems. This research designs and implements a Smart Parcel Box based on face recognition using a Convolutional Neural Network as a package security solution in boarding house areas. The system integrates a Raspberry Pi 3B+ as the main processing unit, an ESP32 microcontroller, a GM67 barcode sensor for tracking number scanning, an HC-SR04 ultrasonic sensor for box capacity monitoring, and a solenoid door lock as the locking actuator. Face recognition employs a multi-task cascaded convolutional network for face detection and FaceNet on TensorFlow Lite for face recognition, with Cosine Similarity at a threshold of 0.75. An Android mobile application integrated with Firebase manages user data, package tracking numbers, and pickup history in real-time. Test results show 85.8% face recognition accuracy. The GM67 sensor achieves a 92.3% barcode reading success rate and has an optimal reading range of 10–40 cm, while the ultrasonic sensor yielded 99.49% average accuracy with 0.51% average error.  
Code-mixed sentiment analysis of Prabowo-Gibran on social media X using mBERT model Alifia Puspaningrum; Nurul Fauziah; Muhamad Mustamiin; Meyer Mega Eklesia Silaban
JURNAL ELTEK Vol. 24 No. 2 (2026): October 2026
Publisher : Politeknik Negeri Malang

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33795/eltek.v24i2.10852

Abstract

The sociopolitical dynamics following the 2024 Indonesian Presidential Election have significantly increased the intensity of public discourse on social media. Polarized opinions regarding various strategic policies are frequently expressed through Indonesian–English code-mixed text. Such texts are characterized by language switching, lexical variations, and semantic ambiguity, posing a major challenge for conventional Natural Language Processing (NLP) methods to accurately capture linguistic markers. This study evaluates Multilingual Bidirectional Encoder Representations from Transformers (mBERT) for opinion classification on Indonesian–English code-mixed text. mBERT was selected because it is a pre-trained model trained on monolingual corpora of 104 languages, granting it cross-lingual transfer capabilities without the need for explicit supervision or bilingual alignment. Additionally, to address the class imbalance issue in the dataset, this study applies the Synthetic Minority Over-sampling Technique (SMOTE). Experimental results demonstrate that mBERT achieved the best performance compared to other models, reaching an accuracy of 82.26% without SMOTE. Following the application of SMOTE, the accuracy increased to 83.00% (an increase of 0.74 percentage points). Although this overall numerical increase appears moderate, the value holds crucial significance: the model successfully mitigated the bias toward majority class predictions. This 0.74% increase represents a significant improvement in precisely recognizing minority classes, as evidenced by vastly more balanced class-wise evaluation metrics. In conclusion, mBERT is a highly promising approach for code-mixed opinion classification, while SMOTE plays an essential role in ensuring predictive fairness on imbalanced datasets.