The voltage difference needed to cause this difference in fuel level is (48.4-47-57)/187.41=4.4mV, so the measured fluctuation is indeed too low.
But your ADC has 10Bit, i.e. gives 1024 different values. At an ADC range of 5V, one step is 5V/1024=4.883mV, which corresponds to 0.9% fuel level. This is the resolution of your circuit, and a change of voltage by 1mV can cause a change of ADC reading by 1 step, which results in a fuel level change by 0.9%.
(The difference between 4.4mV and 4.883mV can be due to rounding or not exact 5V supply voltage)
An analog low pass filter, as suggested by @Edesign can help to reduce the frequency of changes, but can not eliminate them. And: the jumps are still 0.9%.
Averaging of several ADC measurements is similar to the low pass filter. It also makes the fuel level looking smoother, since it blurs the 0.9% steps.
What you can do to reduce the fluctuation is applying a dead band. Something like
Accept ADC value only when it's different by at least 2 steps from the last accepted values.
However, this results in larger steps of 1.8%.
Looking at your formula, the voltage range going into the ADC is roughly about 0V-0.5V (assuming it can measure up to 100% fuel level), so you use only 1/10 of the range of the ADC. If you can map the input voltage to the full ADC range, you can get a resolution of 0.09%.
Either add an amplifier, or set the upper reference voltage of the ADC to about 0.5V (I don't know if this PIC allows it. Many allow to use voltages provided at two pins as upper/lower ADC range limit)
While you get finer steps with this, you'll also get more noise (fluctuations). But you now already got some ideas how to reduce this.