I would suggest adjusting the connector from Oracle to Apache Kafka.
You can refer to the following links for more details:
Using "numeric.mapping": "best_fit" should help, as long as you have specified the precision and scale in your NUMBER type (e.g., NUMBER(5,0)).