One of the problems in the use of Raman spectroscopy for cancer detection in clinical application is the variety of Raman instruments, producing different spectra for the same sample, due to the nature of the measurement system. This prevents the measured spectra from different systems to be compared against one another without appropriate tools and techniques. Therefore, for each instrument one needs to spend considerable amount of time to prepare a set of reference data based on which the future measurements to be interpreted. For early diagnosis of cancer by Raman spectroscopy, there is a need for an algorithm by which such diagnosis can be made by any type of Raman instrument giving rise to the same findings. In the present study we have investigated the detection of breast cancer in three classes of breast samples (normal, benign and cancer) using three different Raman instruments (Almega, Bruker and R3000) to develop an algorithm that, irrespective of the type of Raman instrument, can be applied to the spectra to extract the features necessary to arrive at the same diagnosis. In doing so, we employed different pre-processing methods to eliminate the instrument-dependent effects on the spectra enabling us to fuse such spectra obtained from different instruments. Then, we classified the data using support vector machine (SVM) and multi-layer perception (MLP) to assess the degree to which the employed methods have been able to detect cancer. The results of the study showed that the range and resolution matching using spline interpolation, and noise and fluorescence elimination using wavelet and SNV normalizations were the most sensitive and accurate procedures for eliminating the instrumental specification-based effects and fusing the data from different instruments.