This paper presents an advanced method for condition monitoring and fault detection in induction motors using a Multiclass Extreme Learning Machine (ELM) classification system and enhanced for feature visualization by t-distributed Stochastic Neighbor Embedding (t-SNE). The approach leverages Motor Current Signature Analysis (MCSA) to identify the characteristic pattern and presence of peaks in the current signal of the motor, indicative of a fault. Three-phase currents from an induction motor powering an industrial belt conveyor are acquired and pre-processed with a Butterworth low-pass filter set at a 100 Hz cut-off frequency, followed by Fast Fourier Transform (FFT) application. This pre-processed data establishes a baseline current signature during normal operation, which is essential for continuous monitoring. Deviations from this baseline are detected using signal processing techniques to identify subtle changes in amplitude, frequency, and phase of the current waveform, signaling potential stator inter-turn faults or rotor bar failures. The features, including the mean, skewness, and variance of the three-phase stator currents, are extracted from time-domain signals and fed into the ELM classifier to determine the type of fault. Visualization with t-SNE assists in identifying distinct clusters corresponding to different fault types. This proposed system provides an effective and efficient solution for real-time fault detection and diagnosis in induction motors, enhancing reliability and operational safety in industrial applications.