We leverage machine learning approaches to adapt nanopore sequencing basecallers for nucleotide modification detection. We first apply the incremental learning technique to improve the basecalling of modification-rich sequences, which are usually of high biological interests. With sequence backbones resolved, we further run anomaly detection on individual nucleotides to determine their modification status. By this means, our pipeline promises the single-molecule, single-nucleotide and sequence context-free detection of modifications. We benchmark the pipeline using control oligos, further apply it in the basecalling of densely-modified yeast tRNAs and E.coli genomic DNAs, the cross-species detection of N6-methyladenosine (m6A) in mammalian mRNAs, and the simultaneous detection of N1-methyladenosine (m1A) and m6A in human mRNAs. Our IL-AD workflow is available at: https://github.com/wangziyuan66/IL-AD.