diff --git a/bugbug/commit_features.py b/bugbug/commit_features.py
index 4d7f3b0bda..f0ac448ca6 100644
--- a/bugbug/commit_features.py
+++ b/bugbug/commit_features.py
@@ -3,6 +3,7 @@
 # License, v. 2.0. If a copy of the MPL was not distributed with this file,
 # You can obtain one at http://mozilla.org/MPL/2.0/.
 import sys
+from pathlib import Path
 from typing import Sequence
 
 import pandas as pd
@@ -813,6 +814,39 @@ def __call__(self, commit, **kwargs):
         return commit["files"]
 
 
+class FilesPathComponents(object):
+    name = "filespathcomponents"
+
+    def __call__(self, commit, **kwargs):
+        return list(
+            {
+                part
+                for path in commit["files"]
+                for part in Path(path).parts
+                if part not in ("/", "")
+            }
+        )
+
+
+class PlatformKeywords(object):
+    def __init__(self, platform, keywords):
+        self.platform = platform
+        self.keywords = [kw.lower() for kw in keywords]
+        self.name = f"{platform} keywords"
+
+    def __call__(self, commit, **kwargs):
+        count = sum(
+            1
+            for path in commit["files"]
+            if any(
+                kw in part.lower() for part in Path(path).parts for kw in self.keywords
+            )
+        )
+        desc_lower = commit["desc"].lower()
+        count += sum(1 for kw in self.keywords if kw in desc_lower)
+        return count
+
+
 def _pass_through_tokenizer(doc):
     return doc
 
@@ -863,6 +897,15 @@ def __call__(self, commit, **kwargs):
         return commit["types"]
 
 
+class TypesCounts(object):
+    def __call__(self, commit, **kwargs):
+        counts: dict[str, int] = {}
+        for path in commit["files"]:
+            type_ = repository.get_type(path)
+            counts[f"{type_} count"] = counts.get(f"{type_} count", 0) + 1
+        return counts
+
+
 def merge_metrics(objects):
     metrics = {}
 
@@ -890,6 +933,7 @@ def merge_commits(commits: Sequence[repository.CommitDict]) -> repository.Commit
     return repository.CommitDict(
         {
             "nodes": list(commit["node"] for commit in commits),
+            "desc": " ".join(commit["desc"] for commit in commits),
             "pushdate": commits[0]["pushdate"],
             "types": list(set(sum((commit["types"] for commit in commits), []))),
             "files": list(set(sum((commit["files"] for commit in commits), []))),
@@ -965,9 +1009,12 @@ def merge_commits(commits: Sequence[repository.CommitDict]) -> repository.Commit
 
 class CommitExtractor(BaseEstimator, TransformerMixin):
     def __init__(self, feature_extractors, cleanup_functions):
-        assert len(set(type(fe) for fe in feature_extractors)) == len(
-            feature_extractors
-        ), "Duplicate Feature Extractors"
+        assert len(
+            set(
+                fe.name if hasattr(fe, "name") else type(fe)
+                for fe in feature_extractors
+            )
+        ) == len(feature_extractors), "Duplicate Feature Extractors"
         self.feature_extractors = feature_extractors
 
         assert len(set(type(cf) for cf in cleanup_functions)) == len(
@@ -1011,7 +1058,7 @@ def transform(self, commits):
                 # FIXME: This is a workaround to pass the value to the
                 # union transformer independently. This will be dropped when we
                 # resolve https://github.com/mozilla/bugbug/issues/3876
-                if isinstance(feature_extractor, Files):
+                if isinstance(feature_extractor, (Files, FilesPathComponents)):
                     result[sys.intern(feature_extractor_name)] = res
                     continue
 
diff --git a/bugbug/model.py b/bugbug/model.py
index 58255f81ed..9616ec19ed 100644
--- a/bugbug/model.py
+++ b/bugbug/model.py
@@ -218,6 +218,8 @@ def get_human_readable_feature_names(self):
                 feature_name = f"Combined text contains '{feature_name}'"
             elif type_ == "files":
                 feature_name = f"File '{feature_name}'"
+            elif type_ == "filespathcomponents":
+                feature_name = f"File path component '{feature_name}'"
             elif type_ not in ("data", "couple_data"):
                 raise ValueError(f"Unexpected feature type for: {full_feature_name}")
 
@@ -226,6 +228,12 @@ def get_human_readable_feature_names(self):
         return cleaned_feature_names
 
     def get_important_features(self, cutoff, shap_values):
+        # In the multi-class case, we have (n_samples, n_features, n_classes) and
+        # we need to normalize it to (n_classes, n_samples, n_features) for the logic
+        # below to work.
+        if isinstance(shap_values, np.ndarray) and shap_values.ndim == 3:
+            shap_values = np.moveaxis(shap_values, -1, 0)
+
         # returns top features for a shap_value matrix
         def get_top_features(cutoff, shap_values):
             # Calculate the values that represent the fraction of the model output variability attributable
diff --git a/bugbug/models/__init__.py b/bugbug/models/__init__.py
index eb1a66a84f..8b5468e78a 100644
--- a/bugbug/models/__init__.py
+++ b/bugbug/models/__init__.py
@@ -33,6 +33,7 @@
     "testlabelselect": "bugbug.models.testselect.TestLabelSelectModel",
     "testgroupselect": "bugbug.models.testselect.TestGroupSelectModel",
     "testconfiggroupselect": "bugbug.models.testselect.TestConfigGroupSelectModel",
+    "testconfig": "bugbug.models.testfailure.TestConfigModel",
     "testfailure": "bugbug.models.testfailure.TestFailureModel",
     "tracking": "bugbug.models.tracking.TrackingModel",
     "uplift": "bugbug.models.uplift.UpliftModel",
diff --git a/bugbug/models/testfailure.py b/bugbug/models/testfailure.py
index 0f84a5121c..ea890de981 100644
--- a/bugbug/models/testfailure.py
+++ b/bugbug/models/testfailure.py
@@ -4,6 +4,7 @@
 # You can obtain one at http://mozilla.org/MPL/2.0/.
 
 import logging
+from typing import Optional
 
 import xgboost
 from imblearn.pipeline import Pipeline as ImblearnPipeline
@@ -12,14 +13,44 @@
 from sklearn.feature_extraction import DictVectorizer
 from sklearn.feature_extraction.text import CountVectorizer
 from sklearn.pipeline import Pipeline
+from sklearn.utils.class_weight import compute_sample_weight
 
-from bugbug import commit_features, repository, test_scheduling, utils
+from bugbug import (
+    commit_features,
+    db,
+    feature_cleanup,
+    repository,
+    test_scheduling,
+    utils,
+)
 from bugbug.model import CommitModel
 
 logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)
 
 
+class _BalancedXGBClassifier(xgboost.XGBClassifier):
+    def fit(self, X, y, **fit_params):
+        fit_params["sample_weight"] = compute_sample_weight("balanced", y)
+        return super().fit(X, y, **fit_params)
+
+
+PLATFORM_KEYWORDS = (
+    (("linux",), "linux"),
+    (("windows", "win"), "windows"),
+    (("android", "apk", "fenix", "focus", "klar"), "android"),
+    (("macosx",), "mac"),
+)
+
+
+def get_platform(task_name: str) -> Optional[str]:
+    config = task_name.split("/")[0]
+    for keywords, platform in PLATFORM_KEYWORDS:
+        if any(k in config for k in keywords):
+            return platform
+    return None
+
+
 class TestFailureModel(CommitModel):
     def __init__(self, lemmatization=False):
         CommitModel.__init__(self, lemmatization)
@@ -36,12 +67,7 @@ def __init__(self, lemmatization=False):
             commit_features.SourceCodeDeleted(),
             commit_features.OtherDeleted(),
             commit_features.TestDeleted(),
-            # commit_features.author_experience(),
-            # commit_features.reviewer_experience(),
             commit_features.ReviewersNum(),
-            # commit_features.component_touched_prev(),
-            # commit_features.directory_touched_prev(),
-            # commit_features.file_touched_prev(),
             commit_features.Types(),
             commit_features.Files(),
             commit_features.Components(),
@@ -74,7 +100,7 @@ def __init__(self, lemmatization=False):
                                 CountVectorizer(
                                     analyzer=utils.keep_as_is,
                                     lowercase=False,
-                                    min_df=0.0014,
+                                    min_df=0.005,
                                 ),
                                 "files",
                             ),
@@ -134,3 +160,205 @@ def get_labels(self):
 
     def get_feature_names(self):
         return self.clf.named_steps["union"].get_feature_names_out()
+
+
+class TestConfigModel(CommitModel):
+    def train_test_split(self, X, y):
+        from sklearn.model_selection import train_test_split
+
+        return train_test_split(X, y, test_size=0.1, random_state=0)
+
+    def __init__(self, lemmatization=False):
+        CommitModel.__init__(self, lemmatization)
+
+        self.training_dbs.append(test_scheduling.TEST_LABEL_SCHEDULING_DB)
+        self.training_dbs.append(test_scheduling.PUSH_DATA_CONFIG_GROUP_DB)
+
+        feature_extractors = [
+            commit_features.SourceCodeFileSize(),
+            commit_features.OtherFileSize(),
+            commit_features.TestFileSize(),
+            commit_features.SourceCodeAdded(),
+            commit_features.OtherAdded(),
+            commit_features.TestAdded(),
+            commit_features.SourceCodeDeleted(),
+            commit_features.OtherDeleted(),
+            commit_features.TestDeleted(),
+            commit_features.Types(),
+            commit_features.TypesCounts(),
+            commit_features.FilesPathComponents(),
+            commit_features.Components(),
+            commit_features.ComponentsModifiedNum(),
+            commit_features.DirectoriesModifiedNum(),
+            commit_features.SourceCodeFilesModifiedNum(),
+            commit_features.OtherFilesModifiedNum(),
+            commit_features.TestFilesModifiedNum(),
+            commit_features.SourceCodeFileMetrics(),
+            commit_features.PlatformKeywords("mac", ["cocoa", "mac"]),
+            commit_features.PlatformKeywords(
+                "windows", ["wmf", "winlauncher", "dxgi", "hresult", "playready"]
+            ),
+            commit_features.PlatformKeywords("linux", ["vulkan", "wayland"]),
+            commit_features.PlatformKeywords("android", ["geckoview", "fenix"]),
+        ]
+
+        cleanup_functions = [
+            feature_cleanup.fileref(),
+            feature_cleanup.url(),
+            feature_cleanup.synonyms(),
+        ]
+
+        self.extraction_pipeline = Pipeline(
+            [
+                (
+                    "commit_extractor",
+                    commit_features.CommitExtractor(
+                        feature_extractors, cleanup_functions
+                    ),
+                ),
+            ]
+        )
+
+        self.clf = ImblearnPipeline(
+            [
+                (
+                    "union",
+                    ColumnTransformer(
+                        [
+                            ("data", DictVectorizer(), "data"),
+                            (
+                                "filespathcomponents",
+                                CountVectorizer(
+                                    analyzer=utils.keep_as_is,
+                                    lowercase=False,
+                                    min_df=0.01,
+                                ),
+                                "filespathcomponents",
+                            ),
+                            (
+                                "desc",
+                                self.text_vectorizer(min_df=0.01, stop_words="english"),
+                                "desc",
+                            ),
+                        ]
+                    ),
+                ),
+                (
+                    "estimator",
+                    _BalancedXGBClassifier(
+                        learning_rate=0.05,
+                        n_estimators=200,
+                        subsample=1.0,
+                        colsample_bytree=0.7,
+                        reg_alpha=0.5,
+                        max_depth=4,
+                        n_jobs=utils.get_physical_cpu_count(),
+                    ),
+                ),
+            ]
+        )
+
+    def _get_label_db_data(self):
+        all_revs = set()
+        platforms = {}
+        for revs, test_datas in test_scheduling.get_test_scheduling_history("label"):
+            rev = revs[0]
+            all_revs.add(rev)
+            has_likely = any(td["is_likely_regression"] for td in test_datas)
+            failing_platforms = set()
+            for td in test_datas:
+                if has_likely:
+                    if not td["is_likely_regression"]:
+                        continue
+                else:
+                    if not td["is_possible_regression"]:
+                        continue
+                if "-test-verify" in td["name"]:
+                    continue
+                platform = get_platform(td["name"])
+                if platform is not None:
+                    failing_platforms.add(platform)
+            if failing_platforms:
+                platforms[rev] = failing_platforms
+        return all_revs, platforms
+
+    def _get_cg_db_data(self):
+        all_revs = set()
+        platforms = {}
+        for revisions, _, _, possible_regressions, likely_regressions in db.read(
+            test_scheduling.PUSH_DATA_CONFIG_GROUP_DB
+        ):
+            rev = revisions[0]
+            all_revs.add(rev)
+            regressions = (
+                likely_regressions if likely_regressions else possible_regressions
+            )
+            failing_platforms = set()
+            for config, group in regressions:
+                platform = get_platform(config)
+                if platform is not None:
+                    failing_platforms.add(platform)
+            if failing_platforms:
+                platforms[rev] = failing_platforms
+        return all_revs, platforms
+
+    def _get_rev_to_revisions(self, classes):
+        rev_to_revisions = {}
+        for revs, _ in test_scheduling.get_test_scheduling_history("label"):
+            if revs[0] in classes:
+                rev_to_revisions[revs[0]] = revs
+        for revisions, _, _, _, _ in db.read(test_scheduling.PUSH_DATA_CONFIG_GROUP_DB):
+            if revisions[0] in classes and revisions[0] not in rev_to_revisions:
+                rev_to_revisions[revisions[0]] = revisions
+        return rev_to_revisions
+
+    def get_labels(self):
+        label_all_revs, label_db_platforms = self._get_label_db_data()
+        cg_all_revs, cg_db_platforms = self._get_cg_db_data()
+
+        both_analyzed = label_all_revs & cg_all_revs
+
+        classes = {}
+        for rev in both_analyzed:
+            platforms = label_db_platforms.get(rev, set()) | cg_db_platforms.get(
+                rev, set()
+            )
+            if not platforms:
+                continue
+            classes[rev] = next(iter(platforms)) if len(platforms) == 1 else "any"
+
+        class_names = sorted(set(classes.values()))
+
+        logger.info("%d pushes considered", len(classes))
+        for label in class_names:
+            logger.info(
+                "%d pushes with label '%s'",
+                sum(1 for lbl in classes.values() if lbl == label),
+                label,
+            )
+
+        return classes, class_names
+
+    def items_gen(self, classes):
+        rev_to_revisions = self._get_rev_to_revisions(classes)
+
+        needed_revisions = set()
+        for revisions in rev_to_revisions.values():
+            needed_revisions.update(revisions)
+
+        commit_map = {}
+        for commit in repository.get_commits():
+            if commit["node"] in needed_revisions:
+                commit_map[commit["node"]] = commit
+
+        assert len(commit_map) > 0
+
+        for rev, revisions in rev_to_revisions.items():
+            commits = tuple(commit_map[r] for r in revisions if r in commit_map)
+            if not commits:
+                continue
+            commit_data = commit_features.merge_commits(commits)
+            yield commit_data, classes[rev]
+
+    def get_feature_names(self):
+        return self.clf.named_steps["union"].get_feature_names_out()
diff --git a/infra/data-pipeline.yml b/infra/data-pipeline.yml
index 78a378900c..4c8c01baee 100644
--- a/infra/data-pipeline.yml
+++ b/infra/data-pipeline.yml
@@ -1263,6 +1263,43 @@ tasks:
         owner: bugbug-team@mozilla.com
         source: ${repository}/raw/master/data-pipeline.yml
 
+    - ID: train-test-config
+      created: { $fromNow: "" }
+      deadline: { $fromNow: "5 days" }
+      expires: { $fromNow: "1 year" }
+      provisionerId: proj-bugbug
+      workerType: compute-large
+      dependencies:
+        - commit-retrieval
+        - test-label-scheduling-history-generator
+        - test-config_group-scheduling-history-push_data-retrieval
+      payload:
+        maxRunTime: 25200
+        image: mozilla/bugbug-base:${version}
+        command:
+          - bugbug-train
+          - testconfig
+
+        artifacts:
+          public/testconfigmodel.tar.zst:
+            path: /testconfigmodel.tar.zst
+            type: file
+          public/metrics.json:
+            path: /metrics.json
+            type: file
+
+      routes:
+        - notify.email.bugbug-team@mozilla.com.on-failed
+        - index.project.bugbug.train_testconfig.${version}
+        - index.project.bugbug.train_testconfig.per_version.${version}.${year}.${month}.${day}.${hour}.${minute}.${second}
+        - index.project.bugbug.train_testconfig.per_date.${year}.${month}.${day}.${hour}.${minute}.${second}.${version}
+        - index.project.bugbug.train_testconfig.latest
+      metadata:
+        name: bugbug train testconfig model
+        description: bugbug train testconfig model
+        owner: bugbug-team@mozilla.com
+        source: ${repository}/raw/master/data-pipeline.yml
+
     - ID: train-needsdiagnosis
       created: { $fromNow: "" }
       deadline: { $fromNow: "3 days" }
@@ -1454,6 +1491,7 @@ tasks:
         - train-test-label-select
         - train-test-group-select
         - train-test-failure
+        - train-test-config
         - train-needsdiagnosis
         - train-accessibility
         - train-performancebug
@@ -1494,6 +1532,7 @@ tasks:
         - train-spambug
         - train-test-label-select
         - train-test-group-select
+        - train-test-config
         - train-needsdiagnosis
         - train-accessibility
         - train-performancebug