The presence of chlorine in municipal solid waste incineration bottom ash is a critical factor restricting its potential for reuse. However, the current understanding of the speciation characteristics and formation mechanisms of chlorine within the ash remains significantly limited. This study systematically investigated the influence of waste composition and incineration parameters on the content and speciation distribution of total chlorine, water-soluble chlorine, and water-insoluble chlorine in bottom ash. This was achieved through experimental simulations involving both the standalone incineration of municipal solid waste and its co-incineration with waste printed circuit boards (PCBs) and two types of sludge (Sludge 1 and Sludge 2). A backpropagation artificial neural network (BPANN) model was developed using the data obtained from these incineration experiments. This model was employed for modeling, training, testing, and ultimately predicting the water-soluble chlorine content to optimize the regulation of chlorine speciation characteristics in the bottom ash. Experimental results revealed that, during the standalone incineration of municipal solid waste, an increased proportion of textile and wood/bamboo components, along with elevated incineration temperatures, contributed to a reduction in the total chlorine content in the bottom ash. Furthermore, both increasing the incineration temperature and extending the residence time led to a higher proportion of water-insoluble chlorine in the ash. In the context of co-incineration, compared to standalone municipal solid waste incineration, co-incineration with waste PCBs at 950 °C resulted in an overall increase in total chlorine content in the bottom ash, with 950 °C identified as the optimal temperature for promoting the formation of water-soluble chlorine in the ash. The addition of Sludge 2 caused a general increase in total chlorine, where the increase in water-insoluble chlorine was more pronounced than that of water-soluble chlorine. Notably, the incorporation of Sludge 1 significantly inhibited the formation of water-soluble chlorine compared to co-incineration with PCBs. The BPANN model was successfully developed to predict the water-soluble chlorine content. Model prediction on the training dataset yielded a fitting coefficient (R²) of 0.74 and an average prediction error of 0.06, indicating a high overall predictive accuracy for the model. Further validation through cross-validation and independent experimental verification confirmed the model′s robust performance. Based on the machine learning predictions, it was determined that under optimized conditions—specifically, a plastic∶paper∶textile ratio of 3∶2∶1, inclusion of 50% waste PCBs, an incineration temperature of 1001 °C, and an incineration time of 10 min—the predicted water-soluble chlorine content in the bottom ash increased significantly compared to the lowest predicted value. This comprehensive study provides a crucial theoretical foundation for developing source control strategies aimed at the efficient removal of chlorine from incineration bottom ash.