Complaints like this usually generate as much noise here as is on the powerline. From X10's documentation for extended commands at
https://www.laser.com/dhouston/X10 xtdcode.pdfData is accepted bit by bit as the presence or absence of 120Khz carrier occurring after the positive or negative mains zero crossing. The acceptance window begins approximately 250 usecs. and ends approximately 900 usecs. after a zero crossing. In this window 48 or more cycle of carrier are accepted as a "1" bit and fewer than 48 as a "0" bit. Except for the Startcode, each bit of data is sent in its true and complement form.
The root of the noise problem is that X10 designers were penny-wise and noise-foolish. They saved pennies by omitting decoupling capacitors and they left unused MCU pins floating. This made switches and modules sensitive to powerline spikes which X10 acknowledged in one of their FAQs (no longer online).
You may see where others have postulated that noise generating devices might actually create valid X10 signals but they do not explain why these noise sources have a preference for ON signals, why they do not show up in CM11A or CM15A logs or how the noise manages to create the alternating 01 and 10 bits in the complex error-correction pattern required by the X10 PLC protocol.
Tuicemen's suggestion of using the CM15A log to determine whether there are stray X10 commands online is the best advice. If no commands are logged for these addresses, you can be certain that you have a noise problem.
Most noise problems are local so a whole-house filter is rarely needed.