Too much errors at too many stages for that project.

First off you must separate precision from resolution. A therminstor has lousy precision. The tolerances are large. The drift is large.

Then you get the precision problems from noise, from ADC resolution, from ADC non-linearity, from ADC voltage reference precision, from inexact reference temperature during calibration, drift in voltage reference, ...

The nicest "cheap" temperature sensors you can use that gives reasonably good resolution and precision aren Dallas/Maxim one-wire sensors. You may have a 1/16C resolution and quite good precision.

A DS18S20 has 0.5C accuracy between -10C and +85C. -55 to +125C full working range.

Similar values for DS18B20 but you get up to 12 bits of resolution directly, without needs to do a bit of trickery.
For extended temperature range you get +/- 2C error. 0.2C drift.

You'll have a hard time to find anything that match the resolution and precision of the Dallas one-wire sensors at that cost.