Update src/streamlit_app.py

src/streamlit_app.py

@@ -391,6 +391,7 @@ with tabs[0]:
     )
     st.markdown(f"Total features loaded: **{df.shape[1]}**  |  Rows: **{df.shape[0]}**")
 
+
 # ----- Visualization tab
 with tabs[1]:
     st.subheader("Feature Visualization")
@@ -414,6 +415,35 @@ with tabs[1]:
         ax2.set_title("Operating Mode Clusters (PCA Projection)")
         st.pyplot(fig2, clear_figure=True)
 
+        # --- PCA Explanation ---
+        st.markdown("""
+        **Interpretation – Operating Mode Clusters**
+
+        This PCA-based projection compresses over 100 process features into two principal dimensions,
+        revealing the dominant patterns in furnace operation. Each color represents an automatically discovered
+        *operating mode* (via K-Means clustering).
+
+        - **Distinct clusters (colors)** → different operating regimes (e.g., high-power melt, refining, tapping, idle)
+        - **Overlaps** → transitional phases or process variability
+        - **Compact clusters** → stable operation; **spread-out clusters** → drift or unstable control
+        - **Shifts between colors** over time may reflect raw-material change or arc power adjustment
+
+        Understanding these clusters helps metallurgists and control engineers associate process signatures
+        with efficient or energy-intensive operating conditions.
+        """)
+
+        # --- Dynamic insight: which features drive PCA the most ---
+        from sklearn.decomposition import PCA
+        num_df = df.select_dtypes(include=[np.number]).fillna(0)
+        pca = PCA(n_components=2, random_state=42)
+        pca.fit(num_df)
+        comp_df = pd.DataFrame(pca.components_.T, index=num_df.columns, columns=["PC1", "PC2"])
+        top_pc1 = comp_df["PC1"].abs().nlargest(5).index.tolist()
+        top_pc2 = comp_df["PC2"].abs().nlargest(5).index.tolist()
+        st.info(f"**Top variables driving PCA-1 (X-axis):** {', '.join(top_pc1)}")
+        st.info(f"**Top variables driving PCA-2 (Y-axis):** {', '.join(top_pc2)}")
+
+
 # ----- Correlations tab
 with tabs[2]:
     st.subheader("Correlation explorer")