Update app.py

app.py

@@ -1,13 +1,15 @@
 from fastapi import FastAPI
 from pydantic import BaseModel
 import numpy as np
+import pandas as pd
 import joblib
 
 # -----------------------------
-# Load model + scaler
+# Load model + scaler + encoder
 # -----------------------------
-model = joblib.load("appliance_model.pkl")
-scaler = joblib.load("scaler.pkl")
+model = joblib.load("xgb_model.pkl")
+scaler = joblib.load("xgb_scaler.pkl")
+le = joblib.load("label_encoder.pkl")
 
 # -----------------------------
 # FastAPI app
@@ -15,7 +17,7 @@ scaler = joblib.load("scaler.pkl")
 app = FastAPI(title="Energy Appliance Detection API")
 
 # -----------------------------
-# Input schema
+# Input schema (RAW FEATURES ONLY)
 # -----------------------------
 class Features(BaseModel):
     mean: float
@@ -23,11 +25,20 @@ class Features(BaseModel):
     min: float
     std: float
     range: float
-    skew: float
-    kurtosis: float
     peak_count: float
     slope: float
 
+# -----------------------------
+# Feature order (VERY IMPORTANT)
+# -----------------------------
+columns = [
+    "mean","max","min","std","range",
+    "peak_count","slope",
+    "energy","ratio",
+    "cv","peak_ratio",
+    "delta_mean","power_density"
+]
+
 # -----------------------------
 # Routes
 # -----------------------------
@@ -39,31 +50,51 @@ def root():
 def health():
     return {"status": "ok"}
 
+# -----------------------------
+# Prediction
+# -----------------------------
 @app.post("/predict")
 def predict(data: Features):
 
-    features = np.array([[
-        data.mean,
-        data.max,
-        data.min,
-        data.std,
-        data.range,
-        data.skew,
-        data.kurtosis,
-        data.peak_count,
-        data.slope
-    ]])
-
-    # Scale
-    features_scaled = scaler.transform(features)
-
-    # Predict
-    pred = model.predict(features_scaled)[0]
-    probs = model.predict_proba(features_scaled)[0]
+    # -------- RAW --------
+    mean = data.mean
+    mx = data.max
+    mn = data.min
+    std = data.std
+    rng = data.range
+    peak = data.peak_count
+    slope = data.slope
+
+    # -------- ENGINEERED FEATURES --------
+    energy = mean * 40
+    ratio = mx / (mn + 1)
+    cv = std / (mean + 1)
+    peak_ratio = peak / 40
+    delta_mean = std * 0.8
+    power_density = mean / (rng + 1)
+
+    row = [
+        mean, mx, mn, std, rng,
+        peak, slope,
+        energy, ratio,
+        cv, peak_ratio,
+        delta_mean, power_density
+    ]
+
+    df = pd.DataFrame([row], columns=columns)
+
+    # -------- SCALE --------
+    X = scaler.transform(df)
+
+    # -------- PREDICT --------
+    probs = model.predict_proba(X)[0]
+    pred_index = np.argmax(probs)
     confidence = float(np.max(probs))
 
-    # Confidence filtering
-    if confidence < 0.45:
+    pred = le.inverse_transform([pred_index])[0]
+
+    # -------- UNKNOWN FILTER --------
+    if confidence < 0.5:
         pred = "Unknown"
 
     return {