The maintenance and inspection of power transformers can be a time-consuming task for electric utilities, but it is a necessity for maintaining electrical grid reliability. A standard strategy for diagnosis of fault conditions in an oil-filled transformer is to periodically acquire oil samples and perform dissolved gas analysis (DGA). Aging and temperature variation can induce varying concentrations of hydrogen, methane, and other hydrocarbons, which all form as the oil degrades. Acetylene (C2H2) is generated only during localized, high-temperature events such as partial discharge, and its presence is a key marker for identifying these conditions..The development of optical fiber-based sensors to fill the role of DGA offers several advantages, including the option to implement real-time, in-situ, or even spatially resolved (distributed or quasi-distributed) sensing schemes. The evanescent field approach, in conjunction with tailored sensing materials, provides a cheap and scalable solution to this problem. However, this solution is oftentimes hampered by long-term stability and cross-sensitivity issues. One solution is to gather data from multiple optical fiber sensors designed to eliminate cross-sensitivity and calibrate drift. In this work, a multi-sensor array is developed to target multiple gas species relevant to transformer monitoring (C2H2, CH4, H2). This approach, combined with machine learning models such as support vector machines (SVM), can be used to identify the gas species present at concentrations relevant to DGA (ppm levels) with dramatically increased accuracy.