My suggestion are:
To check difference between Attempt without HO and Seizures without HO.
To check distributions of measurement on each TRX in DL and UL direction.
To check Idle band (somebody could transmit in your frequency band or any of your NB site)
Very important is just the time, on when the statistic was broken.
To check reason for HO. If necessary than we could run Cell Trace on the problematic cell and also on the neighbor cell. From measurement we could observe, what is wrong (bad level or bad quality).
From my experiences, if you set HO margin to low in one direction and to back (0 dB) there are lot of ping pong HO. We have to set some limit of repetition of PWRBGT HO.
If you set HO RXLEV access to low (-102 dBm) there will be a lot of unsuccessful HO.
Check settings of frequency plan (BCCH, TCH, HSN, MAIO). I saw sometimes that on the site with HSN=XX (synchronized manager) was MAIO=0 on two cells.
And many others.