Detected a *potential* strand bias > 1% in an unstranded protocol

[kumarage]

Read lengths I use are ~150bp.

"fR31=4_1H3_R1_merged_val_1.fq.gz
fR32=4_1H3_R2_merged_val_2.fq.gz

salmon quant -i ${salmon_index}
-l IU -1 $fR31
-2 $fR32
--validateMappings -o transcripts_quant_maprate
--numGibbsSamples 20 --gcBias --seqBias

"
When checked lib_format_counts.json it says:
{
"read_files": "[ 4_1H3_R1_merged_val_1.fq.gz, 4_1H3_R2_merged_val_2.fq.gz]",
"expected_format": "IU",
"compatible_fragment_ratio": 1.0,
"num_compatible_fragments": 4935867,
"num_assigned_fragments": 4935867,
"num_frags_with_concordant_consistent_mappings": 4356425,
"num_frags_with_inconsistent_or_orphan_mappings": 830423,
"strand_mapping_bias": 0.6801021479768388,
"MSF": 0,
"OSF": 0,
"ISF": 2962814,
"MSR": 0,
"OSR": 0,
"ISR": 1393611,
"SF": 384481,
"SR": 445942,
"MU": 0,
"OU": 0,
"IU": 0,
"U": 0
}

I get a low mapping rate ~25%. Is there any way to improve?