The Internet of Things (IoT) enables the next economic revolution in which data and immediacy are the key players. Edge computing is a compelling alternative for enabling computing capabilities at the network's edge. These computing capabilities could help transform the generated data into useful information by executing machine learning (ML) workloads. TinyML is emerging as a fast-growing ML ecosystem field aiming to perform workloads on typically battery-operated devices, providing sensor data analytics at extremely low power (mW). In this work, we evaluate a wide range of TinyML and edge computing platforms to assess the computational/energy trade-off of these platforms. In this work, we analyse the Arduino Nano 33 BLE Sense microcontroller and three different edge computing devices, namely the CPU-based Raspberry Pi 4 and the CPU-GPU Nvidia Jetson platforms Nano and AGX Xavier. We run two lightweight artificial neural networks (ANN) to forecast the internal temperature of an operational greenhouse. Our results show that the microcontroller-based devices can offer a competitive and energy-efficient computational alternative to more traditional edge computing approaches for lightweight ML workloads.