Density fluctuations and the Widom line are of great importance in understanding the critical phenomena and the behaviors of supercritical fluids (SCFs). We report on the direct classification of liquid-like and gas-like molecules coexisting in the SCF, identified by machine learning analysis on simulation data. The deltoid coexistence region encloses the Widom line and may therefore be termed the Widom delta. Number fractions of gas-like and liquid-like particles are found to undergo continuous transition across the delta, following a simplified two-state model. These fractions are closely related to the magnitude of supercritical anomaly, which originates from the fluctuation between the two types. This suggests a microscopic view of the SCF as a mixture of liquid-like and gas-like structures, providing an integrative explanation to the anomalous behaviors near the critical point and the Widom line.