This paper presents the development and analysis of a sensor system based on multiplexed intensity variation sensors using a polymer optical fiber (POF). The sensor is based on a multiplexing technique from side-coupling of light emitting diodes (LEDs) with the sequential activation of the light sources. In this case, 20 sensors are developed from the side-coupling between the lateral section of the POF and each LED. The sensor system is embedded in a polydimethylsiloxane (PDMS) resin for encapsulation, chemical and mechanical protection of the sensors. The sensors sensitivities as a function of the refractive index and pressure are experimentally obtained. Then, the sensors are positioned inside an acrylic tank for the liquid level measurement and the water and oil identification. The tests using water and oil mixtures in the tank are performed to evaluate the possibility of measuring water and oil levels. To that extent, an ensemble learning algorithm using random forest (RF) is applied on the responses of the 20 sensors to obtain the level of each fluid. In addition, the algorithm's parameters of the RF approach are optimized using the minimization of root mean squared error (RMSE) as the objective. The dataset are divided into train and test samples with an additional dataset used as validation of the proposed RF-based regression model. Results show the feasibility of the proposed sensor system, where the mean errors for water and total levels are 0.14 cm and 0.08 cm, respectively. In the validation tests, a determination coefficient (R2) of 0.99 is obtained and a RMSE of 3.51 cm and 2.69 cm for water and total levels, respectively.