In neuroimaging study, a challenging issue is elucidation of natural brain processing in a less-controlled environment. Thanks to the advancement in technologies to miniaturize the sensor device (EEG, NIRS, etc.) and wireless communication, portable/wearable devices are now available, allowing participants to be released from tight experimental conditions. To examine "natural behavior" signatures based on such portable devices, however, there are two main problems, i) how to analyze recorded signals with external/internal artifacts in the real environment, ii) how to detect behavior events from the signals without cue stimulus. To this end, we developed a wearable multi-channel fNIRS-EEG system (kNIRS-EEG). In the unique experimental environment that mimics a living house, called BMI house, we developed a system to record NIRS-EEG and environmental data simultaneously. We examined NIRS-EEG recordings from both single-participant and dual-participants simultaneous sessions. The participants were instructed to perform sequential daily-life and free moving behaviors. In off-line analysis, we labeled each behavior a tag and registered it to a "Brain-log database". After each segment of 8ch NIRS signals was associated with the behavioral tag that signifies the class of daily-life behaviors (i.e. operating TV/AC, reading etc.), we attempted to decode each behavior from the NIRS signals. We applied SVM for the decoding and evaluated timing of the decodable period. Our preliminary analysis found that the classification accuracy rose after the onset of the behavior segment. This result suggested that we could extract some behavior-related information from the recorded signal even if it contains motion artifacts. It is plausible to decode a participant's action and/or intention based on database that associates the brain signals with behavior labels determined from environmental sensors.