Chemoresistance is a major obstacle to effective cancer treatment, particularly in tubo-ovarian high-grade serous carcinoma (HGSC), the most lethal gynaecological malignancy. Predicting patient-specific chemoresistance remains challenging due to tumour heterogeneity and the lack of reliable biomarkers. Raman spectroscopy, a label-free technique that provides biochemical insights into cells and tissues without the need for specific biomarkers, has been extensively applied in cancer research, but its full potential for detecting subtle biochemical changes linked to chemoresistance in HGSC at the single-cell and subcellular levels remains underexplored. Another critical challenge is the estimation of classification performance on future data with cross-validation (CV) in the presence of batch effects. In this study, we demonstrated that confocal Raman microscopy combined with multivariate analysis can discriminate between cisplatin-resistant (TYK-nu-CP.r) and cisplatin-sensitive (TYK-nu) HGSC cell lines with 78% accuracy without batch correction. After batch correction, the accuracy improved to 84%. Feature importance analysis suggested that the separation was linked to a higher level of lipid unsaturation and elevated glutathione levels in the chemosensitive cell line. Additionally, we proposed a new CV-based area under the receiver operating characteristic curve (AUC) estimator that accounts for the batch effects better than the popularly used leave-one-batch-out CV. Together, these results show that with careful data processing, accounting for biases and batch effects, Raman microscopy enables reliable detection of chemoresistance at a cellular level and can provide insights into the molecular basis of chemoresistance. This study suggests that Raman microscopy holds promise as a tool for predicting chemoresistance in HGSC and guiding personalised treatment strategies.