Recently, new energy-efficient routing algorithms have been developed for multipath multicast routing to find a stable path for data transmission between nodes that is reliable and fault-tolerant. These algorithms aim to reduce the number of failed data transmissions and the time lost due to route disconnection. To improve data delivery and reduce data loss, several existing techniques have been developed for multipath routing. However, the proposed work focuses on MANETs, which are networks with a variable number of mobile nodes scattered randomly throughout the network. The learning model is first intended to determine which neighbouring nodes are most efficient for data transmission by estimating the link quality and data depletion rate. This leads to increased data packet delivery in a shorter time frame. The weak learners are then subjected to a stacking-based SVM classification technique to produce correct classification results for each node. The degree of resemblance between several data packets is assessed using the Jaccard resemblance function. The studies involve a number of performance measures, including end-to-end delay, the packet delivery ratio, normalized control overhead, energy computing, and residual energy. They are carried out using the ns2.34 simulator. The simulation findings show that the suggested stacking-based SVM model enhances the longevity and reliability of MANETs’ data supply, with end delays ranging from 4.659606 ms to 13.610754 ms for node speeds ranging from 10 m/s to 100 m/s.