Optimized quality of service prediction in FSO links over South Africa using ensemble learning

Free Space Optical (FSO) communication is considered effective and efficient for high-speed wireless data transmission, particularly in urban environments. However, the quality of service (QoS) in FSO links is highly susceptible to environmental factors such as atmospheric turbulence, rain, fog, and dust storms, which are common in South Africa. The current demands for high-capacity free-space optical (FSO) communications, in terms of transmission capacity, latency, and reliability, are driven by rapid advancements in communication technology and the massive utilization of data. However, FSO communications are significantly affected by adverse weather conditions such as fog, rain, snow, smoke, dust, and drizzle, all of which impact the Quality of Service (QoS) provided by these systems. This article aims to optimize the QoS using ensemble learning models such as Random Forest (RF), AdaBoost Regressor (ABR), Stacking Regressor (SR), Gradient Boosting Regressor (GBR), and Multilayer Neural Network (MLNN). To achieve this goal, meteorological data, including visibility, wind speed, and altitude, were obtained from the archives of the South African Weather Service over a period of ten years (2010–2019) at four locations: Polokwane, Kimberley, Bloemfontein, and George. We estimated the data rate, received power, fog-induced attenuation, bit error rate (BER), and power penalty using the collected and processed data. For an optical attenuation of approximately 1 dB/km at a wavelength of 1550 nm across the study locations—Polokwane, Kimberley, Bloemfontein, and George—the transmitted power of the FSO link is capable of supporting data rates of 1.62 × 10¹⁴, 7.68 × 10¹², 2.80 × 10¹², and 2.90 × 10¹³ bps, respectively. At the same wavelength (1550 nm) and a visibility of 1 km, the values of the attenuation factor across Polokwane, Kimberley, Bloemfontein, and George are 2.13, 2.13, 2.13, and 2.13 dB/km, respectively. Furthermore, for attenuation of 1 dB/km, the BER values across the study locations—Polokwane, Kimberley, Bloemfontein, and George—are 6.743 × 10⁻¹³, 6.730 × 10⁻¹³, 6.750 × 10⁻¹³, and 6.755 × 10⁻¹³, respectively. The RMSE and R-squared values of the SR model across all the study locations, Polokwane, Kimberley, Bloemfontein, and George, are 0.0073 and 0.9951, 0.0065 and 0.9998, 0.0060 and 0.9941, and 0.0032 and 0.9906, respectively. The result showed that using ensemble learning techniques in transmission modeling can significantly enhance service quality and meet customer service level agreements. Regardless of the linked climate vicissitude along the transmission connection, the outcomes demonstrated that the ensemble method was successful in efficiently optimizing the signal-to-noise ratio, which in turn enhanced the QoS at the point of reception.