Fixed effects vs. random effects in plm

I have unbalanced panel data, with different observations on different companies over a 10 year period. However, when choosing whether to use Fixed effects or Random effects I did a Hausman test to decide, and the p value vas < 0.05 so fixed effect is preferable. However, R adjusted is negative when using fixed effects and positive (and good) when using random effects? Is there any way where we can use the random effects despite the test, or is there a way of improving the R adjusted in the fixed effects model?