It　is　of　great　importance　to　categorize　the　number　of　bugs　over　time　for　both　software　project　managers　and　its　end　users.　This　paper　proposes　a　novel　approach　called　BugCat　(i.e.,　Bug　number　Categorization)　to　categorize　the　bug　numbers　of　software　with　multi-modal　time　series　learning.　The　time　series　derived　from　the　five　modalities　are　used　as　the　inputs　of　the　proposed　BugCat　approach　as　the　bug　number,　the　day　of　the　week,　the　day　off,　bug　severity　and　bug　priority.　Then,　the　LSTM　(Long　Short-Term　Memory)　embedding　is　conducted　on　the　five　modalities　of　times　series　separately　and,　the　concatenated　vectors　derived　from　data　fusion　on　the　five　LSTM　embeddings　are　used　as　the　input　of　the　full-connected　neural　network　with　ReLU　(Rectified　Linear　Unit)　activation　to　categorize　the　bug　numbers　of　software.　The　extensive　experiments　with　the　Mozilla　Firefox　bug　data　demonstrate　the　superiority　of　the　proposed　BugCat　approach　over　state-of-the-art　techniques　including　multi-layer　perceptron　(MLP),　fully　convolutional　network　(FCN)　and　LSTM.　©　2022,　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.