Globally, early childhood caries (ECC) is a significant public health concern, necessitating effective prediction and prevention strategies. This study aimed to explore variations in the oral microbiome of saliva from pre-school Han and Uyghur children during ECC development and establish a predictive model based on temporal oral microbiome changes. Saliva samples were collected from a single kindergarten every three months over six months. Forty-four pre-school children provided 132 samples, categorized into six groups: (1) HEF (healthy pre-school Han children), (2) HEO (Han children with caries), (3) HEP (Han children with progressive caries), (4) WEF (healthy pre-school Uyghur children), (5) WEO (Uyghur children with caries), and (6) WEP (Uyghur children with progressive caries). Illumina Miseq sequencing identified oral microbiome differences between groups and time points. The Random Forest (RF) algorithm established ECC prediction models. The T1HEO group exhibited significantly higher Chaol index, observed species index, PD whole tree index, and Shannon index than the T2HEO group (p < 0.01). Similarly, the T1WEO group had significantly higher Chaol index, observed species index, and PD whole tree index than the T2WEO group (p < 0.05). The AUROC value for the ECC prediction model based on temporal oral flora changes was 0.517 (95% CI: 0.275-0.759) for pre-school Han children and 0.896 (95% CI: 0.78-1.00) for pre-school Uyghur children. In the onset of caries in pre-school Han children, bacterial species richness and community diversity in saliva declined, paralleled by a decrease in bacterial species richness in pre-school Uyghur children's oral saliva. The ECC prediction model grounded on temporal oral microflora changes exhibited robust predictive power, particularly for pre-school Uyghur children, potentially leading to more effective ECC prevention measures.