Malcolm Low’s September 22, 2026 case report describes training a Titanic-competition ensemble on an Android 14 phone using Termux, a Debian/Ubuntu userspace, Antigravity CLI and Kaggle’s command-line tools. The reported 10-fold blend combined CatBoost, Random Forest and Extra Trees; it reached a 0.80382 Kaggle public score, while its 0.8698 out-of-fold accuracy and 0.9029 out-of-fold ROC-AUC are separate validation figures. These are the author’s reported results, not an independently reproduced benchmark.
What ran on the phone—and what ran on Kaggle
In the reported workflow, the Android phone was the local computing host. Termux provided the terminal environment, and the author used PRoot to run a Debian/Ubuntu userspace. Antigravity CLI orchestrated the Python machine-learning code; the Kaggle CLI handled competition interaction. In this setup, Kaggle was the source of competition data and the destination for submissions, not evidence that the model training ran in a Kaggle cloud notebook.
As an Amazon Associate I earn from qualifying purchases.
The author reports using Android 14, Python 3.14, and ARM64 builds of CatBoost, scikit-learn, pandas and NumPy. Termux describes itself as an Android terminal emulator and Linux environment that works without rooting, with packages managed through APT and Python available. Its project documentation also describes Bluetooth keyboard and external-display support. A keyboard can make long terminal sessions more comfortable, but it is optional.
Google describes Antigravity CLI as a terminal-first interface for working with coding agents and executing shell commands. Its official product materials list macOS, Windows, Linux and Googlebook downloads; they do not document support for Android or Termux under PRoot. The Android configuration should therefore be understood as the author’s reported setup, not an officially confirmed Google-supported installation.
#1 Best Overall
- YOUR CONTENT, SUPER SMOOTH: The ultra-clear 6.7" FHD+ Super AMOLED display of Galaxy A17 5G helps bring your content to life, whether you're scrolling through recipes or video chatting with loved ones.¹
- LIVE FAST. CHARGE FASTER: Focus more on the moment and less on your battery percentage with Galaxy A17 5G. Super Fast Charging powers up your battery so you can get back to life sooner.²
- MEMORIES MADE PICTURE PERFECT: Capture every angle in stunning clarity, from wide family photos to close-ups of friends, with the triple-lens camera on Galaxy A17 5G.
- NEED MORE STORAGE? WE HAVE YOU COVERED: With an improved 2TB of expandable storage, Galaxy A17 5G makes it easy to keep cherished photos, videos and important files readily accessible whenever you need them.³
- BUILT TO LAST: With an improved IP54 rating, Galaxy A17 5G is even more durable than before.⁴ It’s built to resist splashes and dust and comes with a stronger yet slimmer Gorilla Glass Victus front and Glass Fiber Reinforced Polymer back.
Local Android execution versus Kaggle notebooks
| Consideration | Phone running Termux | Kaggle-hosted notebook |
|---|---|---|
| Where computation runs | On the Android device in the reported setup. | In Kaggle’s cloud notebook environment. |
| Hardware and packages | Depends on the phone’s hardware and whether the required Android ARM64 packages install and work in the userspace. | Kaggle documents notebook environments with configurable accelerators and attached data sources. |
| Environment and collaboration | The case report gives its software versions, but does not independently establish a reproducible installation path. | Kaggle documents versioned notebook environments and collaboration features. |
| Best fit | Experimenting with local execution when the device and package stack are compatible. | Using Kaggle’s hosted data-science environment and available cloud resources. |
Kaggle’s CLI documentation recommends Python 3.11 or later and describes commands for competitions, datasets and notebooks. That recommendation does not, by itself, verify that every dependency in this particular Python 3.14 ARM64 setup works. The case report’s installation details were not independently reproduced, so treat compatibility as specific to the author’s account rather than a general guarantee.
How the 10-fold validation works
In k-fold cross-validation, the training data is divided into k parts. The model trains on k−1 parts and is evaluated on the remaining part; this repeats until every part has served as the held-out fold. The scores from those held-out predictions are then summarized. With 10 folds, each record is held out once across ten rounds. Scikit-learn notes that cross-validation can be computationally expensive, since the model must be fitted repeatedly.
Rank #2
- Carrier: This phone is locked to Tracfone, which means this device can only be used on the Tracfone wireless network. Tracfone plan required, activating is easy, just 3 steps.
- DISPLAY: Immersive viewing on a 6.7-inch super-bright 120Hz display with powerful stereo speakers and Bass Boost for cinematic entertainment.
- CAMERA SYSTEM: Advanced 50MP Quad Pixel camera captures sharp, detailed photos and videos in any lighting condition
- PERFORMANCE: Lightning-fast 5G connectivity paired with a powerful processor and RAM Boost for smooth multitasking.
- BATTERY LIFE: Long-lasting 5000mAh battery with TurboPower charging technology delivers hours of power in minutes.
An out-of-fold (OOF) prediction is a prediction for a training record made by a model that did not train on that record’s fold. OOF accuracy measures the share of those predictions that match the labels. ROC-AUC measures how well the model ranks positive cases above negative cases across classification thresholds; it is not the same measure as accuracy.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsTen folds do not make an evaluation automatically unbiased or guarantee a stronger model. They describe how the training data was split and evaluated. Repeatedly choosing features or settings based on a public leaderboard can still leak information from that evaluation into development. A genuinely untouched final test set remains useful for assessing generalization.
Rank #3
- YOUR CONTENT, SUPER SMOOTH: The ultra-clear 6.7" FHD+ Super AMOLED display of Galaxy A17 5G helps bring your content to life, whether you're scrolling through recipes or video chatting with loved ones.¹
- LIVE FAST. CHARGE FASTER: Focus more on the moment and less on your battery percentage with Galaxy A17 5G. Super Fast Charging powers up your battery so you can get back to life sooner.²
- MEMORIES MADE PICTURE PERFECT: Capture every angle in stunning clarity, from wide family photos to close-ups of friends, with the triple-lens camera on Galaxy A17 5G.
- NEED MORE STORAGE? WE HAVE YOU COVERED: With an improved 2TB of expandable storage, Galaxy A17 5G makes it easy to keep cherished photos, videos and important files readily accessible whenever you need them.³
- BUILT TO LAST: With an improved IP54 rating, Galaxy A17 5G is even more durable than before.⁴ It’s built to resist splashes and dust and comes with a stronger yet slimmer Gorilla Glass Victus front and Glass Fiber Reinforced Polymer back.
What the reported blend combined
The author describes a fixed weighted blend: each component model contributes probabilities, multiplied by a chosen weight, and the weighted contributions are combined for a prediction. The reported weights and settings were:
| Model | Blend weight | Reported settings |
|---|---|---|
| CatBoost | 60% | Depth 4; L2 regularization 4.0 |
| Random Forest | 25% | Depth 5; minimum leaf size 2 |
| Extra Trees | 15% | Depth 5 |
These are the case report’s configuration details, not settings independently tested or recommended for other datasets. A fixed weighted blend is different from stacking: stacking trains another model to combine base-model outputs, while this blend uses specified weights. Scikit-learn documents both as ensemble approaches.
Rank #4
- PRIVACY DISPLAY: Automatically hide your screen from those beside you. The built-in privacy display can be preset¹ to turn on when receiving notifications, typing passwords, or using specific apps
- TYPE IT IN. TRANSFORM IT FAST: Enhance any shot in seconds on your smartphone by using Photo Assist² with Galaxy AI.³ Add objects, restore details, or apply new styles by simply typing or tapping
- NIGHTS, CAPTURED CLEARLY: From gigs to city lights, record and capture moments after dark with clarity using Nightography so your photos and videos stay crisp and clear on your Samsung Galaxy
- MAKE IT. EDIT IT. SHARE IT: Turn everyday moments into something personal with creative tools built right into your mobile phone, whether it’s a special contact photo, custom wallpaper, an invitation or more⁴
- HELP THAT KEEPS UP: Stay in the moment while Now Nudge with Galaxy AI helps you respond faster and stay organized with smart suggestions⁵ that appear exactly when you need them on your phone
Reported results, with the comparisons kept separate
The case report compares a five-fold CatBoost model with a group-survival feature against the ten-fold three-model blend. The author reports that both had the same OOF accuracy, while the blend had a higher public score in that experiment:
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →| Author-reported experiment | OOF accuracy | Kaggle public score | Reported rank |
|---|---|---|---|
| Five-fold CatBoost plus group-survival feature | 0.8698 | 0.79904 | 492 |
| Ten-fold weighted blend | 0.8698 | 0.80382 | 291 of 10,058 competitors; reported as top 2.89% |
For the ten-fold blend, the author also reports an OOF ROC-AUC of 0.9029. The public score and rank are leaderboard figures; OOF accuracy and ROC-AUC are validation metrics. They describe different evaluations and should not be treated as directly interchangeable.
Best Value
- Carrier: This phone is locked to Tracfone, which means this device can only be used on the Tracfone wireless network. Activating is easy, just 3 steps.
- ACTIVATION Promotion: Includes 1500 min, 1500 texts & 1500 MB Data + add more as you need it
- CAMERA SYSTEM: 50MP Quad Pixel camera. Capture sharper, more vibrant photos day or night with 4x the light sensitivity.
- PERFORMANCE: Blazing-fast Qualcomm performance. Get the speed you need for great entertainment with a Snapdragon 680 processor and 4GB of RAM.
- 64GB built-in storage. Get plenty of room for photos, movies, songs, and apps. Made for US
This is one reported comparison, not proof that ten folds or blending generally improves performance. It does not isolate the effect of fold count from the effect of changing the model mix, and the report does not establish an independently checked leaderboard snapshot date or submission.
Why the higher historical-override score is not the clean result
The same article reports a public score of 0.83253 and rank 98, described as top 0.96%, for a historical passenger override experiment. The author labels that experiment disqualified because it used historical passenger information, then describes a decoupled version that returned to the clean model’s score. This higher result is not the weighted blend’s clean score and should not be presented as such.
The underlying issue is the validation boundary: information about the evaluation cases can make a score look better without demonstrating that a model learned patterns that generalize. Scikit-learn warns that repeatedly adjusting a model based on test-set performance leaks information into development and undermines the test as an estimate of generalization. Here, the leakage label and audit are the author’s account; they are not an external adjudication or an official competition disqualification finding.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
What a reader can reasonably take from the case
- A phone can serve as the host for a reported local machine-learning workflow, but this particular Android, PRoot, Antigravity and ARM64 package combination is not independently verified here.
- Kaggle’s hosted notebooks are a distinct option with documented cloud resources; using Kaggle’s CLI for competition interaction does not mean training happened in the cloud.
- Cross-validation helps estimate performance on held-out portions of training data, but neither ten folds nor an ensemble protects against leakage caused by tuning against evaluation results.
- The reported public score belongs to one Titanic experiment. It is not a general promise about Android performance, ensemble quality, or competition ranking.
Sources for the platform descriptions are the Termux project overview, Google’s Antigravity product page, Kaggle’s notebook and CLI documentation, and scikit-learn’s cross-validation and ensemble documentation. Competition figures and implementation specifics are attributed to Malcolm Low’s September 22, 2026 case report.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
