Detail Cantuman

Air penting bagi pengaturan suhu tubuh, transportasi nutrisi, serta fungsi otot, sendi, dan ginjal. Namun banyak orang masih belum mencapai target asupan hariannya. Dalam praktiknya, target asupan cairan tidak seragam antarindividu dan dapat berubah mengikuti kondisi fisiologis serta lingkungan harian seperti suhu dan kelembapan. Diperlukan pendekatan yang mempelajari pola minum personal dan merespons konteks harian secara langsung. Penelitian ini mengembangkan sistem prediksi asupan air harian yang dipersonalisasi dengan menggabungkan akuisisi data berbasis Internet of Things (IoT), rekayasa fitur, dan algoritma Extreme Gradient Boosting (XGBoost). Data diperoleh dari stasiun botol pintar yang dilengkapi sensor berat dan tag RFID, mencatat setiap peristiwa minum beserta waktunya. Fitur model dibangun dari komponen waktu, frekuensi minum, dan interaksi antar variabel, kemudian digunakan untuk melatih model dengan skema time-series split dan hyperparameter tuning sehingga mampu beradaptasi terhadap pola konsumsi harian masing-masing pengguna. Dari sisi kinerja prediktif, pada evaluasi validasi berbasis time-series split, evaluasi model pada salah satu pengguna menunjukkan bahwa model dapat menjelaskan 84.9% variasi asupan harian, dengan rata-rata selisih absolut sekitar 33 mL per hari, dengan penyimpangan kuadrat rata-rata sekitar 39 mL per hari, dan dengan kesalahan persentase simetris sekitar 1.68%. Sistem ini diintegrasikan ke dalam aplikasi mobile yang, selain menampilkan prediksi dan status hidrasi, juga menyediakan asesmen hidrasi yang dirancang kontekstual dan relevan, dengan menyesuaikan ambang & rekomendasi terhadap suhu dan kelembapan harian. Hasil implementasi menunjukkan bahwa penggabungan prediksi berbasis data personal dengan informasi kontekstual dapat meningkatkan relevansi saran hidrasi dan responsivitas sistem terhadap perubahan kondisi pengguna.

Water is essential for thermoregulation, nutrient transport, and muscle, joint, and kidney function. Yet many people still fall short of their daily intake targets. In practice, the target fluid intake is not uniform across individuals and shifts with physiological conditions and day-to-day environmental factors such as temperature and humidity. A method is therefore needed that learns personal drinking patterns and responds directly to daily context. This study develops a personalized daily water intake prediction system by combining Internet of Things (IoT)-based data acquisition, feature engineering, and the Extreme Gradient Boosting (XGBoost) algorithm. Data were collected from a smart bottle station equipped with a weight sensor and RFID tag, recording each drinking event along with its timestamp. The model features were constructed from temporal components, drinking frequency, and variable interactions, then used to train the model with a time-series split scheme and hyperparameter tuning, enabling adaptation to each user’s daily consumption patterns. In terms of predictive performance, on a time-series validation the model evaluation shows that the model can explained 84.9% of daily intake variance, with a mean absolute error of about 33 mL per day, with a root mean square error of about 39 mL per day, and with a symmetric mean absolute percentage error of about 1.68%. The system is integrated into a mobile application that, in addition to displaying predictions and hydration status, provides a hydration assessment designed to be contextual and relevant by adjusting thresholds and recommendations to daily temperature and humidity. The implementation results indicate that combining personalized data-driven predictions with contextual information can enhance the relevance of hydration recommendations and the system’s responsiveness to changes in user conditions.