With　the　rapid　development　of　automotive　usage　and　maintenance　services,　the　expansion　of　the　automotive　maintenance　data　scale　inevitably　brings　about　abnormal　records,these　records　raise　serious　problems　for　the　automotive　maintenance　industries.　This　paper　proposes　a　new　approach　based　on　the　deep　sparse　auto-encoder　to　deal　with　huge　amounts　of　automotive　maintenance　data　in　reality.　The　failure　description,　maintenance　project,　and　maintenance　parts　in　the　data　are　firstly　vectorized　using　advanced　text　representation　algorithms.　Besides　the　measurement　information　in　the　original　data,　the　obtained　dataset　has　more　actual　semantic　information　to　enhance　the　data　correlation.　The　dataset　is　then　pre-processed　with　normalization　to　improve　the　performance　of　the　learning　model.A　developed　deep　sparse　auto-encoder　(DSAE)　detection　method　is　finally　arranged　for　abnormal　automotive　maintenance　records.　The　first　layer　of　the　deep　sparse　auto-encoder　is　adapted　with　a　sparse　suppression　to　match　the　dataset　features.　The　results　show　that　the　developed　deep　sparse　auto-encoder　model　can　correctly　identify　the　abnormal　records,　which　achieves　accuracy,　precision,　F1-Score,　recall,　and　ROC-AUC　as　99.2%,　99.9%,　97.4%,　99.6%,　and　97.6%,　respectively.　The　proposed　model　is　further　compared　with　other　benchmark　methods,　the　developed　deep　sparse　auto-encoder　model　performs　better　in　terms　of　handling　the　abnormal　automotive　maintenance　data　and　accurate　detection.　©　2024　IEEE.