init model

.gitignore

@@ -0,0 +1,2 @@
+__pycache__
+flagged

app.py

@@ -0,0 +1,185 @@
+import gradio as gr
+import numpy as np
+import joblib
+import lightgbm
+
+import librosa
+import numpy as np
+import tensorflow as tf
+from tensorflow.keras.models import load_model
+import pickle
+from tensorflow.keras.models import Model
+from tensorflow.keras.layers import (
+    Input,
+    Conv1D,
+    LSTM,
+    Bidirectional,
+    Dense,
+    Dropout,
+    Concatenate,
+    Layer,
+    BatchNormalization,
+)
+from tensorflow.keras.utils import get_custom_objects
+from tensorflow.keras.saving import register_keras_serializable
+
+
+# 🔹 Attention Layer
+@register_keras_serializable()
+class Attention(Layer):
+    def __init__(self, **kwargs):
+        super(Attention, self).__init__(**kwargs)
+        self.attention_dense = Dense(1, activation=None)
+
+    def call(self, inputs):
+        attention_scores = self.attention_dense(inputs)
+        attention_weights = tf.nn.softmax(attention_scores, axis=1)
+        return tf.reduce_sum(inputs * attention_weights, axis=1)
+
+
+# ✅ Register the custom object manually
+get_custom_objects()["Attention"] = Attention
+
+
+# 🔹 Normalize Data Properly
+def normalize(X):
+    return (X - np.min(X, axis=0)) / (np.max(X, axis=0) - np.min(X, axis=0) + 1e-8)
+
+
+# Load the trained model
+lr_model = joblib.load("saved_model/best_lgbm_model.pkl")
+with open("saved_model/scaler_xgb_lgbm.pkl", "rb") as s:
+    scaler = pickle.load(s)
+
+# Define the feature list based on the dataset (excluding 'id' and target column)
+feature_names = [
+    "Age",
+    "Cough",
+    "WeightLoss",
+    "Gender_Male",
+    "Gender_Other",
+    "Fever_Yes",
+    "NightSweats_Yes",
+    "ChestPain_Yes",
+    "Hemoptysis_Yes",
+    "Breathlessness_Moderate",
+    "Breathlessness_Severe",
+    "ContactHistory_Yes",
+    "TravelHistory_Yes",
+    "HIVStatus_Positive",
+    "PreviousTB_Yes",
+]
+
+
+# Define the prediction function
+def predict_tuberculosis(*features):
+    try:
+        # Convert inputs to a NumPy array and reshape for model prediction
+        features_array = np.array(features, dtype=float).reshape(1, -1)
+        print(features_array)
+        features_array_scaled = scaler.transform(features_array)
+        print(features_array_scaled)
+        prediction = lr_model.predict(features_array_scaled)
+        print(prediction)
+
+        return (
+            "Tuberculosis Positive" if prediction[0] == 1 else "Tuberculosis Negative"
+        )
+    except Exception as e:
+        return f"Error: {e}"
+
+
+# Load the trained model
+model = load_model("saved_model/best_model.keras")
+
+
+# Function to extract audio features
+def extract_features(audio_path, sr=22050, max_len=157):
+    try:
+        # Load the audio file
+        y, sr = librosa.load(audio_path, sr=sr)
+
+        # Extract features
+        mfcc = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=20)  # (20, time)
+        chroma = librosa.feature.chroma_stft(y=y, sr=sr)  # (12, time)
+        mel = librosa.feature.melspectrogram(y=y, sr=sr, n_mels=128)  # (128, time)
+        spec_contrast = librosa.feature.spectral_contrast(y=y, sr=sr)  # (7, time)
+
+        # Transpose and pad/crop to max_len
+        def pad_or_truncate(feature, target_shape):
+            if feature.shape[1] > target_shape:
+                return feature[:, :target_shape]
+            else:
+                return np.pad(
+                    feature,
+                    ((0, 0), (0, target_shape - feature.shape[1])),
+                    mode="constant",
+                )
+
+        mfcc = pad_or_truncate(mfcc, max_len).T  # (157, 20)
+        chroma = pad_or_truncate(chroma, max_len).T  # (157, 12)
+        mel = pad_or_truncate(mel, max_len).T  # (157, 128)
+        spec_contrast = pad_or_truncate(spec_contrast, max_len).T  # (157, 7)
+
+        # Normalize
+        def normalize(X):
+            return (X - np.min(X)) / (np.max(X) - np.min(X) + 1e-8)
+
+        mfcc = normalize(mfcc)
+        chroma = normalize(chroma)
+        mel = normalize(mel)
+        spec_contrast = normalize(spec_contrast)
+
+        return (
+            np.array([mfcc]),
+            np.array([chroma]),
+            np.array([mel]),
+            np.array([spec_contrast]),
+        )
+
+    except Exception as e:
+        print(f"Error processing audio: {e}")
+        return None, None, None, None
+
+
+# Function to predict from audio
+def predict_from_audio(audio_path):
+    mfcc, chroma, mel, spec = extract_features(audio_path)
+    if mfcc is None:
+        return "Error processing audio file."
+
+    # Predict
+    prediction = model.predict([mfcc, chroma, mel, spec])
+    predicted_class = np.argmax(prediction, axis=1)[0]
+
+    return (
+        "Tuberculosis Detected" if predicted_class == 1 else "No Tuberculosis Detected"
+    )
+
+
+# Create the Gradio interface
+with gr.Blocks() as demo:
+    gr.Markdown("# Tuberculosis Prediction Model")
+
+    with gr.Tab("Predict with Numeric Features"):
+        inputs = [
+            (
+                gr.Number(label=feature)
+                if feature in ["Age", "Cough", "WeightLoss"]
+                else gr.Radio([0, 1], label=feature)
+            )
+            for feature in feature_names
+        ]
+        predict_button = gr.Button("Predict Tuberculosis")
+        output_text = gr.Textbox(label="Prediction Result")
+        predict_button.click(predict_tuberculosis, inputs, output_text)
+
+    with gr.Tab("Predict with Audio File"):
+        audio_input = gr.Audio(type="filepath", label="Upload Audio File")
+        predict_audio_button = gr.Button("Predict from Audio")
+        audio_output = gr.Textbox(label="Audio Prediction Result")
+        predict_audio_button.click(predict_from_audio, audio_input, audio_output)
+
+# Run Gradio
+if __name__ == "__main__":
+    demo.launch()

requirements.txt

@@ -0,0 +1,10 @@
+python-multipart
+gradio
+numpy
+joblib
+scikit-learn
+librosa
+tensorflow
+ffmpeg-python
+xgboost
+lightgbm

saved_model/best_lgbm_model.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1dfc58d78d32d32e18f0e722458dcbb826cbf901fa40b0ab1aad5ed281a1cc42
+size 330094

saved_model/best_model.keras

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5fbafe813834eb3d50fa78ebd17d16a8d590dba030634dea5733687ee7113921
+size 20165288

saved_model/l1_model.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f69c047a778e5ffc75a271025f101bdf9ffed84709a115957b73c9205c8f918c
+size 821

saved_model/lr_model_selected.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:86d550f7fbccb6b8dd871adb1aa2f6e3d5bbecb3be52f243ac8a163b7b833d3d
+size 777

saved_model/rf_model.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9bd92ee375718cca2f8360e69c0f2a190f880ab04ecd9e471bcd3c7d603769d2
+size 37029516

saved_model/rfe_svm.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:31d9bc253d84dcd1cb08139933a6f1b2802753b89a3eeaf61e053cc3f16a50c4
+size 1210

saved_model/scaler_xgb_lgbm.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:476cbd66b7c4eaf18047ab84aab05abda62a31779e9007f49451d6c976687d81
+size 1144

saved_model/xgb_model.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b4b8c92dea1d682612c703eb023561857254e5f8f66ec48df4089071c3a1532
+size 776