Based on the output provided by infer_experiment.py, the strandedness can be inferred as follows:

• Fraction of reads failed to determine: 0.0795 This indicates the fraction of reads for which the strandedness could not be determined. It could be due to various reasons such as low-quality reads or issues with the library preparation. This value is relatively small, indicating that a majority of the reads could be properly assigned a strandedness.

• Fraction of reads explained by "++,--": 0.0703 This fraction represents the reads that align in the forward orientation ("++") or the reverse orientation ("--"). In other words, the reads are mapping to the same strand as their corresponding reference transcripts. In this case, it suggests that 7.03% of the reads are aligned in the forward orientation and 7.03% are aligned in the reverse orientation.

• Fraction of reads explained by "+-,-+": 0.8502 This fraction represents the reads that align in the forward-reverse ("+-") or reverse-forward ("-+") orientation. In other words, the reads are mapping to the opposite strand compared to their corresponding reference transcripts. In this case, it suggests that 85.02% of the reads are aligned in the forward-reverse orientation ("-+") and 85.02% are aligned in the reverse-forward orientation ("+-").

Based on this information, we can conclude that the library appears to be stranded in a reverse manner. The majority of the reads align to the reverse-complementary strand compared to their corresponding reference transcripts, while a smaller fraction aligns to the same strand.

The majority (85%) of your reads falls in the case described by:

+-,-+
read mapped to ‘+’ strand indicates parental gene on ‘-‘ strand
read mapped to ‘-‘ strand indicates parental gene on ‘+’ strand
  

So yes, your library is reverse stranded. As featureCounts is very fast, you can quickly confirm this by running it with both settings (-s 1 and -s 2), the correct setting will have a much lower count for NoFeature than the incorrect one.