Due　to　the　limited　energy,　memory　space　and　processing　ability　on　wireless　sensor　nodes,　the　batch　learning　method　will　be　infeasible　for　larger　number　of　samples　or　sequence　samples.　This　paper　focuses　on　the　incremental　learning　method　for　kernel　machine　by　involving　L1　regularized,　a　novel　incremental　learning　algorithm　for　L1　regularized　Kernel　Minimum　Squared　Error　machine　(L1-KMSE-Increm)　is　proposed　and　evaluated　on　both　synthetic　and　real　data　sets.　The　simulation　results　reveal　that　L1-KMSE-Increm　algorithm　can　obtain　almost　the　same　prediction　accuracy　as　that　of　corresponding　batch　learning　method,　and　significantly　outperforms　the　competitor　on　the　sparse　ratio　of　model　and　the　running　time.