Abstract— The increase in the amount of data acquired from the monitoring of power system components has motivated utilities to employ effective strategies for processing the information collected. Hence, salient features can be identified and efficient decisions is made. An important component of any power system is power transformers, which have the single highest value of the equipment installed in high-voltage substations. For this reason, significant attention has been devoted to transformer monitoring and diagnostic techniques, resulting in huge volumes of raw data, especially related to the detection of any abnormal transformer behavior. The application of many monitoring tests is therefore not always useful, creating a critical need for a rational method of minimizing the number of monitoring tests without losing essential information about the actual condition of the transformer. This paper presents a statistical approach for evaluating the state of the transformer using machine learning technique. Demonstration of the use of classifier ensemble to predict transformer condition was also made. Index Terms—Transformer condition assessment, classification, ensemble classifier. I. INTRODUCTION Power transformers are among the most expensive equipment of the electric power transmission and distribution system and their condition monitoring is important for the uninterrupted and reliable functioning of the power grid [1]. Compromising up to 60%, power transformers have the single highest value of the equipment installed in high-voltage substation [2]. Therefore, additional attention is currently being paid to life-cycle management and condition-monitoring techniques of transformers because of their important contribution in minimizing maintenance costs, and extending the nominal end of life. In a transformer, especially in-service older transformers, gradual deterioration occurs for a variety of reasons: overloading, lack of maintenance, design problems, environment temperature, and other factors that speed up the deterioration process and reduce life expectancy. Condition monitoring and assessment procedures are implemented with the goal of tracking component behavior and detecting early faults. As a result, maintenance programs can be improved, transformer failure rate can be decreased, direct and indirect costs due transformer outages can also be minimized, while overall system reliability and efficiency is enhanced. However, transformer monitoring techniques and maintenance procedures that produce a set of raw data for evaluating the health of the transformer are dependent on the experimental measurement practices employed and physical features of the asset. Approximately 27 assessment and monitoring methods have been reported in the literature [3], with a number of different techniques being introduced for diagnosing the condition of the transformer. An approach proposed in [4], [2], [5], [6] is based on a health index (HI) that incorporates the majority of thermal, mechanical, electrical, and chemical diagnostic tests. In [7], [8] chemical diagnostic methods and their interpretation schemes were reviewed. The use of techniques based on polarization measurements along with furan analysis for diagnosing the insulating system was presented in [9], [10]. In [11], a transformer state assessment based on the association rule of data and the variable weight synthesizing theory of factor space was proposed. High- performance liquid chromatography (HPLC) was used in these measurements in order to analyze cellulose aging. Other researchers have examined the thermal effect on insulating paper, as reported in [12], [13]. The investigation reported in [14] involved the use of a feature selection technique (FST) with a support vector machine (SVM) for identifying the most informative subset of oil characteristics for transformer condition assessment. The application of artificial intelligence for the estimation of transformers condition are widely investigated in literature based on the monitoring data [15], [16]. A multi-attribute decision- making evaluation model for transformer condition assessment has been introduced in [17]. In [18], artificial neural networks (ANN) and adaptive neuro-fuzzy inference system were proposed to determine the health index for power transformers, where technical and economical parameters are used as an input for the model. A support vector machine (SVM) algorithm has been developed in [19] to provide an intelligent tool for automatic measurement data analysis and transformer condition assessment prediction. SVM, neural networks, fuzzy logic, and particle swarm optimization have all been employed in order to translate test results into condition index, which have been applied to only a subset of monitoring techniques. Also, most of the monitoring techniques studied were capable of providing an assessment of the condition of the transformer, none addressed the number of tests required in order to determine the state of the transformer. This paper demonstrates the application of classifiers ensemble for assessing power transformer condition. Weak Application of Ensemble Classification Method for Power Transformers Condition Assessment Ayman Othman, Student Member, IEEE; Monsef Tahir, Student Member, IEEE; Ramadan El Shatshat, Member, IEEE; Khaled Shaban, Senior Member, IEEE.