Dataset of Radish Microgreens

This dataset was collected from an Internet of Things (IoT)-based environmental monitoring system developed for radish microgreens (Raphanus sativus) cultivation as part of a bachelor's thesis research at Universitas Bina Nusantara, Bandung, Indonesia. The research focused on designing and evaluating an anomaly detection system using the Isolation Forest algorithm within a cloud-centric architecture.

Data were acquired continuously from an IoT device consisting of an ESP32 microcontroller integrated with a DHT22 temperature and humidity sensor and a Capacitive Soil Moisture Sensor V1.2. Sensor readings were transmitted to a cloud backend in near real-time via the MQTT protocol. The three primary environmental parameters recorded are air temperature (°C), air humidity (%), and growing-medium moisture (%).

This repository contains four dataset files. The Raw Dataset of Radish Microgreens contains the complete unprocessed sensor readings collected throughout the monitoring period and consists of 33,852 records.

Before the modelling stage, the raw dataset was split into training and evaluation subsets using an 80:20 ratio. As a result, the Radish Microgreens Training Dataset contains 27,081 records, representing 80% of the original dataset. This dataset subsequently underwent a data cleansing process to remove invalid and implausible sensor readings before being used to train the Isolation Forest model.

The remaining 20% of the raw dataset were allocated to the evaluation subset, resulting in 6,771 records. At this initial stage, the evaluation dataset did not contain any ground-truth labels. To enable quantitative evaluation of the Isolation Forest model, a synthetic anomaly injection process was applied to this subset, whereby a portion of the original sensor readings for temperature, air humidity, and growing-medium moisture were deliberately replaced with out-of-range values to simulate anomalous environmental conditions in radish microgreens cultivation. Following the injection process, each record was assigned a ground-truth label of either "Data Normal" or "Data Anomaly". The resulting labeled dataset serves as the reference ground truth for evaluating the Isolation Forest model using performance metrics such as accuracy, precision, recall, and F1 score.

This dataset is intended to support reproducible research in the fields of IoT-based urban farming, unsupervised anomaly detection, and environmental sensor data analytics.